MLPerfHPCResultsVisualizer.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# MLPerf HPC v0.7 results visualization"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This notebook creates a set of tables and plots to understand MLPerf HPC v0.7 submission results from the publicly available repository.\n",
    "\n",
    "For this purpose, it first parses submission results according to the benchmark application chosen below (make sure they are available under the path referenced in `mlperf_results_paths.json`). Then it defines plotting functions to be used later. The results are then presented in a set of tables as well as figures (at the end of the notebook) that give a visual overview of data staging, algorithimic convergence and system throughput performance."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Choose benchmark"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from collections import namedtuple\n",
    "MLPerfBenchmark = namedtuple('MLPerfBenchmark',['round','name'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Choose benchmark to visualize\n",
    "\n",
    "# benchmark = MLPerfBenchmark(round='hpc_results_v0.7', name='cosmoflow') \n",
    "benchmark = MLPerfBenchmark(round='hpc_results_v0.7', name='deepcam')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# paths of results repos\n",
    "import json\n",
    "with open('mlperf_results_paths.json', 'r') as results_paths_f:\n",
    "    mlperf_results_dirs = json.load(results_paths_f)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Parse results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "import glob\n",
    "import re\n",
    "import os\n",
    "import subprocess\n",
    "import math\n",
    "import functools\n",
    "import inspect\n",
    "import numpy as np\n",
    "from pprint import pprint\n",
    "from IPython.display import display"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'../hpc_results_v0.7/'"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "benchmark_results_dir = mlperf_results_dirs[benchmark.round]\n",
    "mlperf_results_dirs[benchmark.round]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['../hpc_results_v0.7/Fujitsu/results/abci_1024xV100_pytorch_closed/deepcam',\n",
       " '../hpc_results_v0.7/Fujitsu/results/abci_1024xV100_fjpytorch_open/deepcam',\n",
       " '../hpc_results_v0.7/LBNL/results/corigpu_n64_pt1.6.0/deepcam']"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "benchmark_submissions = glob.glob( os.path.join(benchmark_results_dir, '*/results/*/' + benchmark.name) )\n",
    "benchmark_submissions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# %%bash\n",
    "# ls $(pwd)/${benchmark_results_dir}/*/systems/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "mlperf_opts = {\n",
    "    'hpc_results_v0.7': {},\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# MLPerf HPC Cosmoflow hyperparameters\n",
    "mlperf_opts['hpc_results_v0.7']['cosmoflow'] = {\n",
    "\n",
    "    # global_batch_size\n",
    "\n",
    "    # opt_name must be SGD for closed\n",
    "\n",
    "    'opt_opts_names' : dict(\n",
    "        SGD=\\\n",
    "        ['sgd_opt_momentum',\n",
    "         'opt_base_learning_rate',\n",
    "         'opt_learning_rate_warmup_epochs',\n",
    "         'opt_learning_rate_warmup_factor',\n",
    "         'opt_learning_rate_decay_boundary_epochs',\n",
    "         'opt_learning_rate_decay_factor']\n",
    "    ),\n",
    "    'runs_to_submit' : 10,\n",
    "    'eval_target_key': 'eval_error',\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# MLPerf HPC DeepCAM hyperparameters\n",
    "\n",
    "mlperf_opts['hpc_results_v0.7']['deepcam'] = {\n",
    "\n",
    "    # global_batch_size\n",
    "\n",
    "    # opt_name must be AdamW or LAMB  for closed\n",
    "\n",
    "    'opt_opts_names' : dict(\n",
    "        AdamW=\\\n",
    "        ['opt_epsilon',\n",
    "         'opt_base_learning_rate',\n",
    "         'opt_learning_rate_warmup_steps',\n",
    "         'opt_learning_rate_warmup_factor',\n",
    "         'opt_learning_rate_decay_steps',\n",
    "         'opt_learning_rate_decay_factor',\n",
    "         'opt_weight_decay',\n",
    "        ],\n",
    "        LAMB=\\\n",
    "        ['opt_epsilon',\n",
    "         'opt_base_learning_rate',\n",
    "         'opt_learning_rate_warmup_steps',\n",
    "         'opt_learning_rate_warmup_factor',\n",
    "         'opt_weight_decay',\n",
    "        ]\n",
    "#         'validation_frequency', fixed to 100\n",
    "#         'loss_weight_pow'       fixed to -0.125\n",
    "    ),\n",
    "    \n",
    "    'runs_to_submit' : 5,\n",
    "    'eval_target_key': 'eval_accuracy',\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "template_pattern = r'[\\s\\S]*:::MLLOG([\\s\\S]+?\"key\":[\\s]*?\"{}\"[\\s\\S]+?)$[\\s\\S]*'\n",
    "\n",
    "train_samples_pattern =  re.compile(template_pattern.format('train_samples'), re.M) # [\\s\\S]*^ at the beginning not possible due to NVIDIA submission\n",
    "eval_samples_pattern  =  re.compile(template_pattern.format('eval_samples'),  re.M) # [\\s\\S]*^ at the beginning not possible due to NVIDIA submission\n",
    "\n",
    "epoch_start_pattern =  re.compile(template_pattern.format('epoch_start'), re.M) # [\\s\\S]*^ at the beginning not possible due to NVIDIA submission\n",
    "eval_start_pattern  =  re.compile(template_pattern.format('eval_start' ), re.M) # [\\s\\S]*^ at the beginning not possible due to NVIDIA submission\n",
    "eval_stop_pattern   =  re.compile(template_pattern.format('eval_stop'  ), re.M) # [\\s\\S]*^ at the beginning not possible due to NVIDIA submission\n",
    "eval_target_pattern   =  re.compile(template_pattern.format(mlperf_opts[benchmark.round][benchmark.name]['eval_target_key']), re.M) # [\\s\\S]*^ at the beginning not possible due to NVIDIA submission\n",
    "epoch_stop_pattern  =  re.compile(template_pattern.format('epoch_stop' ), re.M) # [\\s\\S]*^ at the beginning not possible due to NVIDIA submission\n",
    "\n",
    "if benchmark.round == 'hpc_results_v0.7':\n",
    "    staging_start_pattern  =  re.compile(template_pattern.format('staging_start' ), re.M) # [\\s\\S]*^ at the beginning not possible due to NVIDIA submission\n",
    "    staging_stop_pattern   =  re.compile(template_pattern.format('staging_stop'  ), re.M) # [\\s\\S]*^ at the beginning not possible due to NVIDIA submission\n",
    "    \n",
    "\n",
    "# Optimizer/hyperparameters\n",
    "global_batch_size_pattern =  re.compile(r'[\\s\\S]*:::MLLOG([\\s\\S]+?\"key\":[\\s]*?\"{}\"[\\s\\S]+?)$[\\s\\S]*'.format('global_batch_size'), re.M) # [\\s\\S]*^ at the beginning not possible due to NVIDIA submission\n",
    "opt_name_pattern =           re.compile(r'[\\s\\S]*:::MLLOG([\\s\\S]+?\"key\":[\\s]*?\"{}\"[\\s\\S]+?)$[\\s\\S]*'.format('opt_name'), re.M)\n",
    "\n",
    "opt_patterns = { \n",
    "    opt_name:  { k: re.compile(r'[\\s\\S]*:::MLLOG([\\s\\S]+?\"key\":[\\s]*?\"{}\"[\\s\\S]+?)$[\\s\\S]*'.format(k), re.M) \n",
    "              for k in mlperf_opts[benchmark.round][benchmark.name]['opt_opts_names'][opt_name] }\n",
    "    for opt_name in mlperf_opts[benchmark.round][benchmark.name]['opt_opts_names']\n",
    "}\n",
    "\n",
    "# Epochs to convergence\n",
    "run_start_pattern =        re.compile(r'[\\s\\S]*^:::MLLOG([\\s\\S]+?\"key\":[\\s]*?\"{}\"[\\s\\S]+?)$'.format('run_start'), re.M)\n",
    "run_stop_pattern =         re.compile(r'[\\s\\S]*^:::MLLOG([\\s\\S]+?\"key\":[\\s]*?\"{}\"[\\s\\S]+?)$'.format('run_stop'), re.M)\n",
    "\n",
    "last_eval_target_pattern = re.compile(r'[\\s\\S]*^:::MLLOG([\\s\\S]+?\"key\":[\\s]*?\"{}\"[\\s\\S]+?)$[\\s\\S]+?\"key\":[\\s]*?\"{}\"[\\s\\S]+'.format(mlperf_opts[benchmark.round][benchmark.name]['eval_target_key'], 'run_stop'), re.M)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "../hpc_results_v0.7/Fujitsu/results/abci_1024xV100_pytorch_closed/deepcam\n",
      "../hpc_results_v0.7/Fujitsu/results/abci_1024xV100_fjpytorch_open/deepcam\n",
      "../hpc_results_v0.7/LBNL/results/corigpu_n64_pt1.6.0/deepcam\n",
      "Staging: No entry in corigpu_n64_pt1.6.0.\n",
      "Staging: No entry in corigpu_n64_pt1.6.0.\n",
      "Staging: No entry in corigpu_n64_pt1.6.0.\n",
      "Staging: No entry in corigpu_n64_pt1.6.0.\n",
      "Staging: No entry in corigpu_n64_pt1.6.0.\n",
      "...done!\n"
     ]
    }
   ],
   "source": [
    "submissions_stats = dict()\n",
    "for submission in benchmark_submissions: \n",
    "\n",
    "    print(submission, flush=True)\n",
    "    submitter = submission.split('/')[-4]\n",
    "    system_name  = submission.split('/')[-2]\n",
    "    with open(os.path.join('/'.join(submission.split('/')[:-3]), 'systems', system_name + '.json'), 'r') as system_desc_file:\n",
    "        system_desc = json.load(system_desc_file)\n",
    "        system_desc['division'] = system_desc['division'].lower()\n",
    "        if not system_desc['division'] in ['open', 'closed']:\n",
    "            print(f'Erroneous division {system_desc[\"division\"]} in system description of {submitter}!')\n",
    "    submission_results = glob.glob(submission + '/result*')\n",
    "    results = []\n",
    "    for result in submission_results:\n",
    "       \n",
    "        # epoch times\n",
    "        epoch_start_content = subprocess.run(['/bin/grep', 'epoch_start', result], capture_output=True).stdout.decode('utf-8')\n",
    "        epoch_starts = [json.loads(epoch_start_pattern.search(l).group(1)) for l in epoch_start_content.split('\\n') if len(l) > 0]\n",
    "\n",
    "        epoch_stop_content = subprocess.run(['/bin/grep', 'epoch_stop', result], capture_output=True).stdout.decode('utf-8')\n",
    "        epoch_stops = [json.loads(epoch_stop_pattern.search(l).group(1)) for l in epoch_stop_content.split('\\n') if len(l) > 0]\n",
    "\n",
    "        if not( len(epoch_starts) == len(epoch_stops) ):\n",
    "            print(f\"Skipping {result} due to non-closed epoch interval...\")\n",
    "            continue\n",
    "        for i in range(len(epoch_starts)):\n",
    "            assert epoch_starts[i]['metadata']['epoch_num'] == epoch_stops[i]['metadata']['epoch_num']\n",
    "\n",
    "        epochs = list(zip([ts['time_ms']/1000. if 'time_ms' in ts else float(ts['time_s']) for ts in epoch_starts], [ts['time_ms']/1000. if 'time_ms' in ts else float(ts['time_s']) for ts in epoch_stops]))\n",
    "\n",
    "        # eval times\n",
    "        eval_start_content = subprocess.run(['/bin/grep', 'eval_start', result], capture_output=True).stdout.decode('utf-8')\n",
    "        eval_starts = [json.loads(eval_start_pattern.search(l).group(1)) for l in eval_start_content.split('\\n') if len(l) > 0]\n",
    "\n",
    "        eval_stop_content = subprocess.run(['/bin/grep', 'eval_stop', result], capture_output=True).stdout.decode('utf-8')\n",
    "        eval_stops = [json.loads(eval_stop_pattern.search(l).group(1)) for l in eval_stop_content.split('\\n') if len(l) > 0]\n",
    "\n",
    "        eval_target_content = subprocess.run(['/bin/grep', mlperf_opts[benchmark.round][benchmark.name]['eval_target_key'], result], capture_output=True).stdout.decode('utf-8')\n",
    "        eval_targets = [json.loads(eval_target_pattern.search(l).group(1)) for l in eval_target_content.split('\\n') if len(l) > 0]\n",
    "        \n",
    "        if not( len(eval_starts) == len(eval_stops) ):\n",
    "            print(f\"Skipping {result} due to non-closed eval interval.\")\n",
    "            continue\n",
    "        elif not(len(eval_starts) in [0, len(eval_targets)]):\n",
    "            print(f\"Skipping {result} due to more with eval_start/eval_stop intervals ({len(eval_starts)}) than {mlperf_opts[benchmark.round][benchmark.name]['eval_target_key']} ({len(eval_targets)}) entries.\")\n",
    "            continue\n",
    "        for i in range(len(eval_starts)):\n",
    "            if 'epoch_num' in eval_starts[i]['metadata']:\n",
    "                assert eval_starts[i]['metadata']['epoch_num'] == eval_stops[i]['metadata']['epoch_num']\n",
    "\n",
    "        if len(eval_starts) == 0 and len(eval_stops) == 0:\n",
    "            print(f\"Warning: {result} has no eval_start/eval_stop logs - assuming training took the entire epoch time.\")\n",
    "            \n",
    "        evals = list(zip([ts['time_ms']/1000. if 'time_ms' in ts else float(ts['time_s']) for ts in eval_starts], [ts['time_ms']/1000. if 'time_ms' in ts else float(ts['time_s']) for ts in eval_stops]))\n",
    "\n",
    "        eval_targets = [(ts['value'], int(ts['metadata']['epoch_num'])) for ts in eval_targets]\n",
    "        \n",
    "        if len(evals) > 0: # not met by some benchmark.round == 'training_results_v0.7' submissions\n",
    "            eval_lb, eval_ub = zip(*evals)\n",
    "            eval_time = np.sum(np.array(eval_ub) - np.array(eval_lb))\n",
    "        else:\n",
    "            eval_time = None\n",
    "\n",
    "        if len(epochs) > 0: # not met by some benchmark.round == 'training_results_v0.7' submissions\n",
    "            epoch_lb, epoch_ub = zip(*epochs)\n",
    "            epoch_time = np.sum(np.array(epoch_ub) - np.array(epoch_lb))\n",
    "        else:\n",
    "            epoch_time = None\n",
    "\n",
    "        train_time = epoch_time - eval_time if eval_time is not None else epoch_time\n",
    "\n",
    "        # staging time\n",
    "        if benchmark.round == 'hpc_results_v0.7':\n",
    "            staging_start_content = subprocess.run(['/bin/grep', 'staging_start', result], capture_output=True).stdout.decode('utf-8')\n",
    "            staging_starts = [json.loads(staging_start_pattern.search(l).group(1)) for l in staging_start_content.split('\\n') if len(l) > 0]\n",
    "\n",
    "            staging_stop_content = subprocess.run(['/bin/grep', 'staging_stop', result], capture_output=True).stdout.decode('utf-8')\n",
    "            staging_stops = [json.loads(staging_stop_pattern.search(l).group(1)) for l in staging_stop_content.split('\\n') if len(l) > 0]\n",
    "            \n",
    "            assert len(staging_starts) == len(staging_stops)\n",
    "            assert len(staging_starts) <= 1\n",
    "                        \n",
    "            if len(staging_starts) == 1:\n",
    "                if system_desc['system_name'] in ['ABCI PRIMERGY CX2570 M4', \n",
    "                                                  'fugaku_16384xA64FX_tensorflow_open', \n",
    "                                                  'fugaku_8192xA64FX_tensorflow_closed',\n",
    "                                                  'ABCI PRIMERGY CX2570 M4',\n",
    "                                                  'fugaku_512xA64FX_tensorflow_closed',\n",
    "                                                  'corigpu_n64_tf1.15.0']:\n",
    "                    run_staging_time = staging_stops[0]['time_ms']/1000. - staging_starts[0]['time_ms']/1000.\n",
    "                else:\n",
    "#                     system_desc['system_name'] in ['coriknl_n512_tf1.15.2',\n",
    "#                                                   'coriknl_n1024_tf1.15.2',\n",
    "#                                                   'daint_gpu_n256_tf2.2.0',\n",
    "#                                                    'daint_gpu_n128_tf2.2.0',\n",
    "#                                                    'Frontera-RTX']\n",
    "                      print(f\"Staging: excluding {system_desc['system_name']}.\")\n",
    "                      run_staging_time = None\n",
    "            else: \n",
    "#                 system_desc['system_name'] in ['hal_v100_n16_tf1.15.0',\n",
    "#                                                'corigpu_n64_pt1.6.0']\n",
    "                print(f\"Staging: No entry in {system_desc['system_name']}.\")\n",
    "                run_staging_time = None\n",
    "        else:\n",
    "            run_staging_time = None\n",
    "            \n",
    "        \n",
    "        # Sample numbers\n",
    "        train_samples_content = subprocess.run(['/bin/grep', 'train_samples', result], capture_output=True).stdout.decode('utf-8')\n",
    "        train_samples = json.loads(train_samples_pattern.search(train_samples_content).group(1))['value']\n",
    "\n",
    "        eval_samples_content = subprocess.run(['/bin/grep', 'eval_samples', result], capture_output=True).stdout.decode('utf-8')\n",
    "        eval_samples = json.loads(eval_samples_pattern.search(eval_samples_content).group(1))['value']        \n",
    "        \n",
    "        # Optimizer parameters\n",
    "        global_batch_size_content = subprocess.run(['/bin/grep', 'global_batch_size', result], capture_output=True).stdout.decode('utf-8')\n",
    "        global_batch_size = json.loads(global_batch_size_pattern.search(global_batch_size_content).group(1))['value']\n",
    "\n",
    "        opt_name_content = subprocess.run(['/bin/grep', 'opt_name', result], capture_output=True).stdout.decode('utf-8')\n",
    "        opt_name = json.loads(opt_name_pattern.search(opt_name_content).group(1))['value']\n",
    "        \n",
    "        if benchmark.name == 'cosmoflow':\n",
    "            if opt_name == 'SGDWD' and submission.endswith('Fujitsu/results/fugaku_8192xA64FX_tensorflow_closed/cosmoflow'):\n",
    "                opt_name = 'SGD'\n",
    "        \n",
    "        opt_opts = {}\n",
    "        if system_desc['division'] == 'closed':\n",
    "            for opt_opt, opt_opt_pattern in opt_patterns[opt_name].items():\n",
    "                if opt_opt == 'sgd_opt_momentum': # no one is using it - forgotten in the reference implementation?\n",
    "                    continue \n",
    "                if benchmark.name == 'deepcam' and opt_opt == 'opt_weight_decay' and 'corigpu_n64_pt1.6.0' in submission:\n",
    "                    opt_opts[opt_opt] = 0.01\n",
    "                    continue                 \n",
    "                opt_opt_content = subprocess.run(['/bin/grep', opt_opt, result], capture_output=True).stdout.decode('utf-8')\n",
    "                opt_opts[opt_opt] = json.loads(opt_opt_pattern.search(opt_opt_content).group(1))['value']\n",
    "\n",
    "\n",
    "        run_stop_content = subprocess.run(['/bin/grep', '-C', '8', 'run_stop', result], capture_output=True).stdout.decode('utf-8')\n",
    "        \n",
    "        run_stop_match = run_stop_pattern.search(run_stop_content)\n",
    "        if run_stop_match is None:\n",
    "            print(f\"Skipping {result} (run_stop not found)\", flush=True)\n",
    "            continue\n",
    "        run_stop = json.loads(run_stop_match.group(1))\n",
    "        if 'status' in run_stop['metadata'] and run_stop['metadata']['status'] != 'success':\n",
    "            print(f\"Skipping {result} (run_stop status was {json.loads(run_stop_match.group(1))['metadata']['status']} != success)\", flush=True)\n",
    "            continue\n",
    "        elif not 'status' in run_stop['metadata']:\n",
    "            print(f\"Warning: {result} has no status field: run_stop['metadata'] = {json.loads(run_stop_match.group(1))['metadata']}\", flush=True)\n",
    "            \n",
    "        run_start_content = subprocess.run(['/bin/grep', 'run_start', result], capture_output=True).stdout.decode('utf-8')\n",
    "        run_start_match = run_start_pattern.search(run_start_content)\n",
    "        run_start = json.loads(run_start_match.group(1))\n",
    "        \n",
    "        run_time = (run_stop['time_ms']-run_start['time_ms'])/1000. if 'time_ms' in run_stop else run_stop['time_s']-run_start['time_s']\n",
    "\n",
    "        last_eval_target_match = json.loads(last_eval_target_pattern.search(run_stop_content).group(1))\n",
    "        run_stop_epoch = last_eval_target_match['metadata']['epoch_num']\n",
    "        \n",
    "        if benchmark.round == 'hpc_results_v0.7' and epoch_starts[0]['metadata']['epoch_num'] == 0:\n",
    "                run_stop_epoch += 1\n",
    "\n",
    "        if benchmark.round == 'hpc_results_v0.7':\n",
    "            run_num_epochs = len(epochs)\n",
    "            run_num_evals = len(evals)\n",
    "        \n",
    "        if run_stop_epoch != run_num_epochs: # epoch counting must start from 0 or 1\n",
    "            if len(epoch_starts) > 0:\n",
    "                raise RuntimeError(f\"Not expected to run {run_num_epochs-run_stop_epoch} extra epoch(s) after reaching target.\")\n",
    "            else:\n",
    "                print(f\"Warning: Ran {run_num_epochs - run_stop_epoch} extra epoch(s) in final block after reaching target.\")\n",
    "\n",
    "        if benchmark.round == 'hpc_results_v0.7' and benchmark.name == 'cosmoflow': # review\n",
    "            if 'fugaku_512xA64FX_tensorflow_closed' in submission:\n",
    "                submission_epochs = [85, 106, 107, 96, 115, 99, 88, 116, 120, 93]\n",
    "                assert run_stop_epoch in submission_epochs\n",
    "                submission_epochs = np.mean(sorted(np.array(submission_epochs))[1:-1])\n",
    "                reference_epochs = np.mean(sorted(np.array([94, 113, 100, 113, 127, 100, 106, 134, 91, 150]))[1:-1])\n",
    "                epochs_adjustment = reference_epochs/submission_epochs\n",
    "                run_time = run_staging_time + (run_time - run_staging_time)*epochs_adjustment\n",
    "                run_stop_epoch *= epochs_adjustment\n",
    "                run_num_epochs *= epochs_adjustment\n",
    "                run_num_evals *= epochs_adjustment\n",
    "                epoch_time *= epochs_adjustment\n",
    "                train_time *= epochs_adjustment\n",
    "                eval_time *= epochs_adjustment                \n",
    "                \n",
    "        results.append( dict(\n",
    "            global_batch_size=global_batch_size, \n",
    "            opt_name=opt_name,\n",
    "            opt_opts=opt_opts,\n",
    "            run_stop_epoch=run_stop_epoch,\n",
    "            run_time=run_time,\n",
    "            run_epoch_throughput=run_stop_epoch/epoch_time,\n",
    "            run_epoch_time=epoch_time/run_stop_epoch,\n",
    "            train_samples=train_samples,\n",
    "            eval_samples=eval_samples,\n",
    "            run_train_time=train_time,\n",
    "            run_eval_time=eval_time,\n",
    "            run_avg_epoch_samples=int(train_samples + (run_num_evals/float(run_num_epochs))*eval_samples),\n",
    "#             run_num_epochs=run_num_epochs, #(same as run_stop_epochs)\n",
    "            run_num_evals=run_num_evals,            \n",
    "            run_train_throughput=run_num_epochs*float(train_samples)/train_time,\n",
    "            run_eval_throughput=run_num_evals*float(eval_samples)/eval_time if eval_time is not None else None,\n",
    "            run_avg_epoch_sample_throughput=run_stop_epoch/epoch_time*int(train_samples + (run_num_evals/float(run_num_epochs))*eval_samples),\n",
    "            run_staging_time=run_staging_time,\n",
    "            run_staging_throughput=float(train_samples + eval_samples)/run_staging_time if run_staging_time is not None else None,\n",
    "#             run_epochs=epochs,\n",
    "#             run_evals=evals,\n",
    "            division=system_desc['division'],\n",
    "            system=system_desc,\n",
    "        ) )\n",
    "    \n",
    "    # Reduce shared features across runs\n",
    "    submission_stats_results = dict()\n",
    "    for k in results[0].keys():\n",
    "        if not k.startswith('run'):\n",
    "            for r in results[1:]:\n",
    "                assert results[0][k] == r[k]\n",
    "            submission_stats_results[k] = results[0][k]\n",
    "    \n",
    "    # run-specific features as a sequence\n",
    "    submission_stats_results['runs'] = []\n",
    "    for r in results:\n",
    "        submission_stats_results['runs'].append({ k: r[k] for k in results[0].keys() if k.startswith('run')})\n",
    "    submissions_stats[submission] = submission_stats_results\n",
    "    \n",
    "def accelerator_number(system):\n",
    "    number_of_nodes = int(system['number_of_nodes'])\n",
    "    accelerators_per_node = int(system['accelerators_per_node'] if int(system['accelerators_per_node']) > 0 else system['host_processors_per_node'])\n",
    "    return number_of_nodes*accelerators_per_node\n",
    "\n",
    "submissions_ids = {k.split('/')[-4]: 0 for k in submissions_stats.keys()}\n",
    "\n",
    "for k in submissions_stats.keys():\n",
    "    submitter = k.split('/')[-4]\n",
    "    system = k.split('/')[-2]\n",
    "    submissions_ids[submitter] += 1\n",
    "    submissions_stats[k]['submission_label'] =       submitter + f'-{submissions_ids[submitter]}/' + system\n",
    "    if benchmark.round != 'hpc_results_v0.7':\n",
    "        submissions_stats[k]['submission_short_label'] = submitter + f'-{submissions_ids[submitter]}'\n",
    "    else:\n",
    "        hpc_v07_systems = {'ABCI':         ['ABCI',         'ABCI',             ], \n",
    "                           'fugaku':       ['Fugaku',       'Fugaku\\n',         ], \n",
    "                           'corigpu':      ['Cori-GPU',     'Cori-GPU\\n',       ], \n",
    "                           'coriknl':      ['Cori-KNL',     'Cori-KNL\\n',       ], \n",
    "                           'daint':        ['Piz-Daint',    'Piz-Daint\\n',      ], \n",
    "                           'hal':          ['HAL',          'HAL',              ], \n",
    "                           'Frontera-RTX': ['Frontera\\n-RTX', 'Frontera\\n-RTX' ]}\n",
    "        for system_name in hpc_v07_systems:\n",
    "            if system_name in submissions_stats[k]['system']['system_name']:\n",
    "                submissions_stats[k]['submission_short_label'] = f'{hpc_v07_systems[system_name][1]}-{accelerator_number(submissions_stats[k][\"system\"])}'\n",
    "                submissions_stats[k]['submission_long_label' ] = f'{hpc_v07_systems[system_name][0]}-{accelerator_number(submissions_stats[k][\"system\"])}'\n",
    "                epochs_system_label = hpc_v07_systems[system_name][0]\n",
    "                if submissions_stats[k]['division'] == 'open':\n",
    "                    epochs_system_label += '\\n'\n",
    "                elif system_name == 'corigpu':\n",
    "                    epochs_system_label = epochs_system_label.replace('-','\\n-')\n",
    "                submissions_stats[k]['submission_epochs_label' ] = f'{epochs_system_label}-{accelerator_number(submissions_stats[k][\"system\"])}'\n",
    "\n",
    "print('...done!')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "../hpc_results_v0.7/Fujitsu/results/abci_1024xV100_pytorch_closed/deepcam\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{'global_batch_size': 2048,\n",
       " 'opt_name': 'LAMB',\n",
       " 'opt_opts': {'opt_epsilon': 1e-08,\n",
       "  'opt_base_learning_rate': 0.0055,\n",
       "  'opt_learning_rate_warmup_steps': 400,\n",
       "  'opt_learning_rate_warmup_factor': 1.0,\n",
       "  'opt_weight_decay': 0.01},\n",
       " 'train_samples': 121266,\n",
       " 'eval_samples': 15158,\n",
       " 'division': 'closed',\n",
       " 'system': {'submitter': 'Fujitsu',\n",
       "  'division': 'closed',\n",
       "  'system_type': 'datacenter',\n",
       "  'status': 'available',\n",
       "  'system_name': 'ABCI PRIMERGY CX2570 M4',\n",
       "  'number_of_nodes': '256',\n",
       "  'host_processors_per_node': '2',\n",
       "  'host_processor_model_name': 'Intel Xeon Gold 6148',\n",
       "  'host_processor_core_count': '40',\n",
       "  'host_processor_vcpu_count': '80',\n",
       "  'host_processor_frequency': '2400MHz',\n",
       "  'host_processor_caches': 'L1: 32KB I + 32KB D per core, L2: 1MB I+D per core, L3: 27.5MB I+D per chip',\n",
       "  'host_processor_interconnect': '3x 10.4GT/s UPI',\n",
       "  'host_memory_capacity': '384GB',\n",
       "  'host_storage_type': 'SSD+NVMe',\n",
       "  'host_storage_capacity': '1600GB',\n",
       "  'host_networking': 'N/A',\n",
       "  'host_networking_topology': 'N/A',\n",
       "  'host_memory_configuration': '12x32GB',\n",
       "  'accelerators_per_node': '4',\n",
       "  'accelerator_model_name': 'NVIDIA Tesla V100',\n",
       "  'accelerator_host_interconnect': 'PCIe 3.0 x16',\n",
       "  'accelerator_frequency': '1530MHz',\n",
       "  'accelerator_on-chip_memories': 'L1: 80x 128KB, L2: 6MB per chip',\n",
       "  'accelerator_memory_configuration': 'HBM2',\n",
       "  'accelerator_memory_capacity': '16GB',\n",
       "  'accelerator_interconnect': '6x 25GT/s NVLink',\n",
       "  'accelerator_interconnect_topology': 'Direct',\n",
       "  'cooling': 'Water-cooled',\n",
       "  'hw_notes': 'none',\n",
       "  'framework': 'PyTorch 1.6.0',\n",
       "  'other_software_stack': 'cuda 10.2.89, cudnn, 7.6.5, cublas 10.2.2.89, gcc 7.4.0, openmpi 2.1.6, python 3.7, apex 0.1',\n",
       "  'operating_system': 'CentOS 7.5.1804',\n",
       "  'sw_notes': 'none'},\n",
       " 'runs': [{'run_stop_epoch': 24,\n",
       "   'run_time': 702.29,\n",
       "   'run_epoch_throughput': 0.04209610543850442,\n",
       "   'run_epoch_time': 23.75516664981842,\n",
       "   'run_train_time': 541.9799993038177,\n",
       "   'run_eval_time': 28.14400029182434,\n",
       "   'run_avg_epoch_samples': 130108,\n",
       "   'run_num_evals': 14,\n",
       "   'run_train_throughput': 5369.910335692159,\n",
       "   'run_eval_throughput': 7540.221638700248,\n",
       "   'run_avg_epoch_sample_throughput': 5477.0400863929335,\n",
       "   'run_staging_time': 132.08399987220764,\n",
       "   'run_staging_throughput': 1032.8578793191555},\n",
       "  {'run_stop_epoch': 24,\n",
       "   'run_time': 703.662,\n",
       "   'run_epoch_throughput': 0.04194352297966711,\n",
       "   'run_epoch_time': 23.841583371162415,\n",
       "   'run_train_time': 544.1070005893707,\n",
       "   'run_eval_time': 28.09100031852722,\n",
       "   'run_avg_epoch_samples': 130108,\n",
       "   'run_num_evals': 14,\n",
       "   'run_train_throughput': 5348.918497368172,\n",
       "   'run_eval_throughput': 7554.447958196671,\n",
       "   'run_avg_epoch_sample_throughput': 5457.187887838528,\n",
       "   'run_staging_time': 131.42799997329712,\n",
       "   'run_staging_throughput': 1038.0132089639799},\n",
       "  {'run_stop_epoch': 24,\n",
       "   'run_time': 714.052,\n",
       "   'run_epoch_throughput': 0.042321324646074306,\n",
       "   'run_epoch_time': 23.62875000635783,\n",
       "   'run_train_time': 538.7329995632172,\n",
       "   'run_eval_time': 28.357000589370728,\n",
       "   'run_avg_epoch_samples': 130108,\n",
       "   'run_num_evals': 14,\n",
       "   'run_train_throughput': 5402.275342998519,\n",
       "   'run_eval_throughput': 7483.58414463429,\n",
       "   'run_avg_epoch_sample_throughput': 5506.342907051436,\n",
       "   'run_staging_time': 146.7960000038147,\n",
       "   'run_staging_throughput': 929.3441237939373},\n",
       "  {'run_stop_epoch': 24,\n",
       "   'run_time': 698.088,\n",
       "   'run_epoch_throughput': 0.04230379413575783,\n",
       "   'run_epoch_time': 23.63854165871938,\n",
       "   'run_train_time': 539.2560000419617,\n",
       "   'run_eval_time': 28.068999767303467,\n",
       "   'run_avg_epoch_samples': 130108,\n",
       "   'run_num_evals': 14,\n",
       "   'run_train_throughput': 5397.035915731177,\n",
       "   'run_eval_throughput': 7560.369153132341,\n",
       "   'run_avg_epoch_sample_throughput': 5504.062047415179,\n",
       "   'run_staging_time': 130.7220001220703,\n",
       "   'run_staging_throughput': 1043.619282696142},\n",
       "  {'run_stop_epoch': 24,\n",
       "   'run_time': 701.077,\n",
       "   'run_epoch_throughput': 0.04218422908559768,\n",
       "   'run_epoch_time': 23.70554166038831,\n",
       "   'run_train_time': 540.8159997463226,\n",
       "   'run_eval_time': 28.117000102996826,\n",
       "   'run_avg_epoch_samples': 130108,\n",
       "   'run_num_evals': 14,\n",
       "   'run_train_throughput': 5381.4680064294635,\n",
       "   'run_eval_throughput': 7547.462361654349,\n",
       "   'run_avg_epoch_sample_throughput': 5488.505677868943,\n",
       "   'run_staging_time': 132.09599995613098,\n",
       "   'run_staging_throughput': 1032.7640507305773}],\n",
       " 'submission_label': 'Fujitsu-1/abci_1024xV100_pytorch_closed',\n",
       " 'submission_short_label': 'ABCI-1024',\n",
       " 'submission_long_label': 'ABCI-1024',\n",
       " 'submission_epochs_label': 'ABCI-1024'}"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# For illustration\n",
    "example_submission = list(submissions_stats.keys())[0]\n",
    "print(example_submission)\n",
    "submissions_stats[example_submission]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Visual analysis: code"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import matplotlib\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "plt.rcParams.update({'font.size': 14})"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Plotting functions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "# kernel density estimator for plot\n",
    "\n",
    "from sklearn.neighbors import KernelDensity\n",
    "from scipy.stats import norm\n",
    "\n",
    "\n",
    "def kde_estimate(run_stop_epochs):\n",
    "    # instantiate and fit the KDE model    \n",
    "    kde = KernelDensity(bandwidth=0.5*np.std(run_stop_epochs), kernel='gaussian')\n",
    "    kde.fit(run_stop_epochs[:,None])\n",
    "\n",
    "    # score_samples returns the log of the probability density\n",
    "    epochs = np.linspace(np.min(run_stop_epochs)-2.5*np.std(run_stop_epochs), np.max(run_stop_epochs)+2.5*np.std(run_stop_epochs), 100)\n",
    "    return epochs, np.exp(kde.score_samples(epochs[:,None]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create level lines in loglog-plots\n",
    "\n",
    "def level_log(value, base):\n",
    "    if base == 10:\n",
    "        return np.log10(value)\n",
    "    elif base == 2:\n",
    "        return np.log2(value)\n",
    "    else:\n",
    "        return np.log(value)/np.log(base)\n",
    "\n",
    "def plot_level_lines(ax, x_range, y_range, level_func, level_label, level_transform=None, level_log_base=10., level_color='black', level_fine_grid=1):\n",
    "    assert level_func in ['x*y', 'y*x', 'x/y', 'y/x']\n",
    "    \n",
    "    # Slopes of lines through corners of plotted window\n",
    "    if level_func in ['x/y', 'y/x']:\n",
    "        min_level =       y_range[0]/x_range[1]\n",
    "        lhs_split_level = y_range[0]/x_range[0]\n",
    "        rhs_split_level = y_range[1]/x_range[1]\n",
    "        max_level =       y_range[1]/x_range[0]\n",
    "    else: # level_func in ['x*y', 'y*x']\n",
    "        min_level =       y_range[0]*x_range[0]\n",
    "        lhs_split_level = y_range[1]*x_range[0]\n",
    "        rhs_split_level = y_range[0]*x_range[1]\n",
    "        max_level =       y_range[1]*x_range[1]\n",
    "\n",
    "    disp_min_level = min_level if level_func != 'x/y' else 1./max_level \n",
    "    disp_max_level = max_level if level_func != 'x/y' else 1./min_level\n",
    "    if level_transform is not None:\n",
    "        disp_min_level = level_transform[0](disp_min_level)\n",
    "        disp_max_level = level_transform[0](disp_max_level)\n",
    "        \n",
    "    log_min_level_ceiled  = np.ceil(level_log(disp_min_level, level_log_base))\n",
    "    log_max_level_floored = np.floor(level_log(disp_max_level, level_log_base))\n",
    "      \n",
    "    coarse_disp_levels = np.logspace(log_min_level_ceiled, log_max_level_floored, int(log_max_level_floored-log_min_level_ceiled)+1, base=level_log_base)\n",
    "    for disp_level in coarse_disp_levels: #[1:-1]:\n",
    "        level = level_transform[1](disp_level) if level_transform is not None else disp_level\n",
    "        level = level if level_func != 'x/y' else 1./level\n",
    "        if level_func in ['x/y', 'y/x']:\n",
    "            if level > lhs_split_level:\n",
    "                lhs = np.array((x_range[0], x_range[0]*level))/2.**0.5\n",
    "            else:\n",
    "                lhs = np.array((y_range[0]/level, y_range[0]))/2.**0.5\n",
    "            if level > rhs_split_level:\n",
    "                rhs = np.array((y_range[1]/level, y_range[1]))*2.**0.5\n",
    "                level_annotate_opts = dict(xytext=(0,5), ha='center', va='bottom') \n",
    "            else:\n",
    "                rhs = np.array((x_range[1], x_range[1]*level))*2.**0.5\n",
    "                level_annotate_opts = dict(xytext=(0,5), ha='center', va='bottom') \n",
    "        else: # level_func in ['x*y', 'y*x']\n",
    "            if level < lhs_split_level:\n",
    "                lhs = np.array((      x_range[0]/2.**0.5, level/x_range[0]*2.**0.5))\n",
    "            else:\n",
    "                lhs = np.array((level/y_range[1]/2.**0.5,       y_range[1]*2.**0.5))\n",
    "            if level < rhs_split_level:\n",
    "                rhs = np.array((level/y_range[0]*2.**0.5,       y_range[0]/2.**0.5))\n",
    "                level_annotate_opts = dict(xytext=(0,-5), ha='center', va='top') \n",
    "            else:\n",
    "                rhs = np.array((      x_range[1]*2.**0.5, level/x_range[1]/2.**0.5))\n",
    "                level_annotate_opts = dict(xytext=(0,-5), ha='center', va='top') \n",
    "        ax.plot([lhs[0], rhs[0]], [lhs[1], rhs[1]], ':', color=level_color, alpha=0.3, zorder=0)\n",
    "        ax.annotate(r\"${log_base}^{{{log_level:d}}}$\".format(log_base=int(level_log_base) if level_log_base == int(level_log_base) else level_log_base, log_level=int(level_log(disp_level,level_log_base))), \n",
    "                    xy=rhs,\n",
    "                    **level_annotate_opts,\n",
    "                    textcoords=\"offset points\", # arrowprops={'width': 1, 'headwidth': 1, 'headlength': 1, 'shrink':0.05, 'color':color },\n",
    "                    color=level_color,\n",
    "                    alpha=0.5)\n",
    "    if level_fine_grid > 1:\n",
    "        assert int(level_log_base) == level_log_base\n",
    "        for lower_coarse_disp_level, upper_coarse_disp_level in zip(coarse_disp_levels[:-1],coarse_disp_levels[1:]):\n",
    "            for disp_level in np.logspace(int(level_log(lower_coarse_disp_level, level_log_base)), int(level_log(upper_coarse_disp_level, level_log_base)), level_fine_grid+1, base=level_log_base)[1:-1]:\n",
    "                level = level_transform[1](disp_level) if level_transform is not None else disp_level\n",
    "                level = level if level_func != 'x/y' else 1./level\n",
    "                if level_func in ['x/y', 'y/x']:\n",
    "                    if level > lhs_split_level:\n",
    "                        lhs = np.array((x_range[0], x_range[0]*level))/2.**0.5\n",
    "                    else:\n",
    "                        lhs = np.array((y_range[0]/level, y_range[0]))/2.**0.5\n",
    "                    if level > rhs_split_level:\n",
    "                        rhs = np.array((y_range[1]/level, y_range[1]))*2.**0.5\n",
    "                    else:\n",
    "                        rhs = np.array((x_range[1], x_range[1]*level))*2.**0.5\n",
    "                else: # level_func in ['x*y', 'y*x']\n",
    "                    if level < lhs_split_level:\n",
    "                        lhs = np.array((      x_range[0]/2.**0.5, level/x_range[0]*2.**0.5))\n",
    "                    else:\n",
    "                        lhs = np.array((level/y_range[1]/2.**0.5,       y_range[1]*2.**0.5))\n",
    "                    if level < rhs_split_level:\n",
    "                        rhs = np.array((level/y_range[0]*2.**0.5,       y_range[0]/2.**0.5))\n",
    "                    else:\n",
    "                        rhs = np.array((      x_range[1]*2.**0.5, level/x_range[1]/2.**0.5))\n",
    "                ax.plot([lhs[0], rhs[0]], [lhs[1], rhs[1]], ':', color=level_color, alpha=0.3, zorder=0)\n",
    "            \n",
    "    ax.annotate(level_label[0],\n",
    "                xy=(x_range[0],y_range[1] if level_func in ['x/y', 'y/x'] else y_range[0]),\n",
    "                xytext=(15,5),\n",
    "                textcoords=\"offset points\",\n",
    "                ha='center', va='bottom',\n",
    "                fontsize='small',\n",
    "                color=level_color,\n",
    "                alpha=0.5)        \n",
    "    ax.annotate(level_label[1],\n",
    "                xy=(x_range[0],y_range[1] if level_func in ['x/y', 'y/x'] else y_range[0]),\n",
    "                xytext=(15,-5),\n",
    "                textcoords=\"offset points\", \n",
    "                ha='center', va='top',\n",
    "                fontsize='small',\n",
    "                color=level_color,\n",
    "                alpha=0.5)    \n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "def create_submission_colors(color_label=None, color_label_range=None, filter_submissions=None, colormap='gist_rainbow'):\n",
    "    cmap = plt.cm.get_cmap(colormap)\n",
    "    colors = cmap(np.arange(0,cmap.N))\n",
    "    if color_label is None:\n",
    "        colors = colors[::len(colors)//len(submissions_stats),:]*0.65\n",
    "        colors = dict(zip(list(submissions_stats.keys()), colors))\n",
    "    else:\n",
    "        if color_label_range is None:\n",
    "            color_label_range = np.unique([color_label(submission_stat) for submission_stat in submissions_stats.values() if filter_submissions(submission_stat)])\n",
    "            print(color_label_range)\n",
    "        colors = colors[::len(colors)//len(color_label_range),:]*0.65\n",
    "        colors = dict(zip(color_label_range, colors))\n",
    "    return colors\n",
    "\n",
    "\n",
    "#Testing\n",
    "def show_submission_colors(color_label, color_label_range=None, filter_submissions=lambda s: True, colormap='gist_rainbow', alpha=1.):\n",
    "    colors = create_submission_colors(color_label=color_label, color_label_range=color_label_range, filter_submissions=lambda s: True, colormap=colormap)\n",
    "    plt.barh([i for i,_ in enumerate(colors)], [1]*len(colors), color=list(colors.values()), tick_label=list(colors.keys()), alpha=alpha)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_all_submissions(x_key, y_key, x_label=None, y_label=None, y_transform=None, y_legend_label=None, y_mean_marker=None, \n",
    "                         secondary_x_key=None, secondary_x_label=None, secondary_x_legend_label=None, secondary_x_marker=None, secondary_x_ticks_same_as_primary=False, primary_to_secondary_x_transform=None, secondary_to_primary_x_transform=None,\n",
    "                         secondary_y_keys=None, secondary_y_axis_label=None, secondary_y_legend_labels=None, secondary_y_transforms=None, secondary_y_markers=None, primary_to_secondary_y_transform=None, secondary_to_primary_y_transform=None,\n",
    "                         title=None, sharey=False, axis_scale='semilogx', filter_submissions=lambda s: True, color_label=None, color_label_range=None, legend_side='right', show_symbolic_legend=False, ax=None,\n",
    "                         level_func=None, level_label=None, level_transform=None, level_color='black', level_log_base=10., level_fine_grid=1, revert_x_axis=False, revert_y_axis=False,\n",
    "                         format_y_axis_as_x_axis=False, submission_annotation_key='submission_short_label', submission_label_annotation_opts=None, color_legend_label=None):\n",
    "    \n",
    "    if type(x_key) is str:\n",
    "        assert not x_key.startswith('run_')\n",
    "    if type(secondary_x_key) is str:\n",
    "        assert not secondary_x_key.startswith('run_')\n",
    "    if type(y_key) is str:\n",
    "        assert y_key.startswith('run_')\n",
    "    if secondary_y_keys is not None:\n",
    "        for secondary_y_key in secondary_y_keys:\n",
    "            if type(secondary_y_key) is str:\n",
    "                assert secondary_y_key.startswith('run_')\n",
    "\n",
    "    fig = None\n",
    "    if ax is None:\n",
    "        fig, ax = plt.subplots(ncols=1, sharey=sharey,  \n",
    "                                figsize=(24, 12))\n",
    "    ax_clsd = ax\n",
    "    ax_open = ax\n",
    "    \n",
    "    colors = create_submission_colors(color_label=color_label, color_label_range=color_label_range, filter_submissions=filter_submissions)\n",
    "\n",
    "    if secondary_x_key is not None:\n",
    "        assert (primary_to_secondary_x_transform is not None) == (secondary_to_primary_x_transform is not None)\n",
    "        secondary_x_axis_functions = (primary_to_secondary_x_transform, secondary_to_primary_x_transform) if (primary_to_secondary_x_transform is not None) and (secondary_to_primary_x_transform is not None) else None\n",
    "        secondary_x_ax_clsd = ax_clsd.secondary_xaxis('top', functions=secondary_x_axis_functions) # FIXME: primary x to secondary x conversion and vice versa tuple for non-identity\n",
    "        if ax_open != ax_clsd:\n",
    "            secondary_x_ax_open = ax_open.secondary_xaxis('top', functions=secondary_x_axis_functions)\n",
    "        else:\n",
    "            secondary_x_ax_open = secondary_x_ax_clsd\n",
    "\n",
    "    if secondary_y_keys is not None:        \n",
    "        assert (primary_to_secondary_y_transform is not None) == (secondary_to_primary_y_transform is not None)\n",
    "        secondary_y_axis_functions = (primary_to_secondary_y_transform, secondary_to_primary_y_transform) if (primary_to_secondary_y_transform is not None) and (secondary_to_primary_y_transform is not None) else None\n",
    "        secondary_y_ax_clsd = ax_clsd.secondary_yaxis('left', functions=secondary_y_axis_functions) # primary x to secondary x conversion and vice versa tuple for non-identity\n",
    "        if ax_open != ax_clsd:\n",
    "            secondary_y_ax_open = ax_open.secondary_yaxis('left', functions=secondary_y_axis_functions)\n",
    "        else:\n",
    "            secondary_y_ax_open = secondary_y_ax_clsd\n",
    "\n",
    "    \n",
    "    x_values_min = dict(closed=float('inf'), open=float('inf')) \n",
    "    x_values_max = dict(closed=float('-inf'), open=float('-inf'))\n",
    "    y_values_min = dict(closed=float('inf'), open=float('inf')) \n",
    "    y_values_max = dict(closed=float('-inf'), open=float('-inf'))\n",
    "    \n",
    "    # Plot all submissions\n",
    "    for i, (k, submission_stat) in enumerate(submissions_stats.items()):\n",
    "        \n",
    "        if not filter_submissions(submission_stat):\n",
    "            continue\n",
    "            \n",
    "        submission_annotation = submission_stat[submission_annotation_key] if type(submission_annotation_key) is str else submission_annotation_key(submission_stat)\n",
    "        if submission_label_annotation_opts is None:\n",
    "            submission_annotation_opts = dict(\n",
    "                        xytext=(-10, 0),\n",
    "                        textcoords=\"offset points\",\n",
    "                        ha='right', va='center',  # rotation=90,\n",
    "                        alpha=1., fontsize=12)\n",
    "        else:\n",
    "            submission_annotation_opts = submission_label_annotation_opts if type(submission_label_annotation_opts) is dict else submission_label_annotation_opts(submission_stat)\n",
    "        color = colors[k if color_label is None else color_label(submission_stat)]\n",
    "            \n",
    "        x_values = submission_stat[x_key] if type(x_key) is str else x_key(submission_stat)\n",
    "        x_values_min[submission_stat['division']] = min(x_values_min[submission_stat['division']], np.min(x_values))\n",
    "        x_values_max[submission_stat['division']] = max(x_values_max[submission_stat['division']], np.max(x_values))\n",
    "        if secondary_x_key is not None:\n",
    "            secondary_x_values = submission_stat[secondary_x_key] if type(secondary_x_key) is str else secondary_x_key(submission_stat)\n",
    "            secondary_x_values = secondary_to_primary_x_transform(secondary_x_values)\n",
    "            x_values_min[submission_stat['division']] = min(x_values_min[submission_stat['division']], np.min(secondary_x_values))\n",
    "            x_values_max[submission_stat['division']] = max(x_values_max[submission_stat['division']], np.max(secondary_x_values))\n",
    "\n",
    "        submission_stat_runs = sorted(submission_stat['runs'], key=lambda run: run['run_time'])\n",
    "        submission_stat_runs = submission_stat_runs[1:mlperf_opts[benchmark.round][benchmark.name]['runs_to_submit']-1] # olympic scoring rules\n",
    "\n",
    "        y_values = np.array([run[y_key] if type(y_key) is str else y_key({'run': run, 'system': submission_stat['system']}) for run in submission_stat_runs])\n",
    "        if secondary_y_keys is not None:\n",
    "            secondary_y_keys_not_none = []\n",
    "            secondary_y_values = []\n",
    "            for secondary_y_key in secondary_y_keys:\n",
    "                secondary_y_values.append( np.array([run[secondary_y_key] if type(secondary_y_key) is str else secondary_y_key({'run': run, 'system': submission_stat['system']}) for run in submission_stat_runs]) )\n",
    "                secondary_y_keys_not_none.append(None not in secondary_y_values[-1])\n",
    "                if secondary_y_keys_not_none[-1]:\n",
    "                    secondary_y_values[-1] = np.array([secondary_to_primary_y_transform(val) for val in secondary_y_values[-1]])\n",
    "\n",
    "        if y_transform is not None:\n",
    "            y_values = np.array([y_transform(val) for val in y_values])\n",
    "        y_values_min[submission_stat['division']] = min(y_values_min[submission_stat['division']], np.min(y_values))\n",
    "        y_values_max[submission_stat['division']] = max(y_values_max[submission_stat['division']], np.max(y_values))\n",
    "\n",
    "        if secondary_y_keys is not None:\n",
    "            for i in np.nonzero(secondary_y_keys_not_none)[0]: \n",
    "                if secondary_y_transforms[i] is not None:\n",
    "                    secondary_y_values[i] = np.array([secondary_y_transforms[i](val) for val in secondary_y_values[i]])\n",
    "                y_values_min[submission_stat['division']] = min(y_values_min[submission_stat['division']], np.min(secondary_y_values[i]))\n",
    "                y_values_max[submission_stat['division']] = max(y_values_max[submission_stat['division']], np.max(secondary_y_values[i]))\n",
    "\n",
    "\n",
    "        if submission_stat['division'] == 'closed':\n",
    "            y_values_unique = np.unique(y_values)\n",
    "            y_values_frequency = np.array([np.sum(e == np.array(y_values)) for e in y_values_unique])\n",
    "            ax_clsd.scatter([x_values]*len(y_values_unique),\n",
    "                        y_values_unique,\n",
    "                        s=5*y_values_frequency,\n",
    "                        color=color,\n",
    "                        label=submission_stat['submission_label'] + (f' ({color_label(submission_stat)})' if color_label is not None else ''),\n",
    "                        alpha=1.)\n",
    "            ax_clsd.annotate(submission_annotation,\n",
    "                        xy=(x_values, np.mean(y_values)),\n",
    "                        color=color,\n",
    "                        **submission_annotation_opts)\n",
    "            if secondary_x_key is not None:\n",
    "                ax_clsd.plot([secondary_x_values, x_values], # requires secondary_x to primary x scale conversion if non-identity\n",
    "                         [np.mean(y_values)]*2,\n",
    "                         '--',\n",
    "                         color=color,\n",
    "                         alpha=0.5)\n",
    "                ax_clsd.scatter([secondary_x_values], # requires secondary_x to primary x scale conversion if non-identity\n",
    "                         [np.mean(y_values)],\n",
    "                         s=25,\n",
    "                         color=color,\n",
    "                         alpha=1.,\n",
    "                         marker=secondary_x_marker, \n",
    "                         facecolors='white')\n",
    "            if secondary_y_keys is not None:\n",
    "                for i in np.nonzero(secondary_y_keys_not_none)[0]: \n",
    "                    x_factor = 2**(0.05*(i+1) if benchmark.name != 'cosmoflow' else 0.1)\n",
    "                    ax_clsd.errorbar([x_values/x_factor],\n",
    "                                     [np.mean(secondary_y_values[i])],\n",
    "                                yerr=[np.std(secondary_y_values[i])],\n",
    "                                fmt=secondary_y_markers[i], \n",
    "                                color=color,\n",
    "                                alpha=1.)\n",
    "                    circle_y, circle_x = horizontal_circle_segment_loglog([np.mean(y_values), np.mean(secondary_y_values[i])], x_values, y_factor=x_factor)\n",
    "                    ax_clsd.plot(circle_x,\n",
    "                             circle_y,\n",
    "                             color=color,\n",
    "                             alpha=.3)\n",
    "            if len(y_values_unique) != 1:\n",
    "                epochs, probs = kde_estimate(y_values)\n",
    "                ax_clsd.plot(   float(x_values)*2**(np.std(y_values)*probs),\n",
    "                            epochs,\n",
    "                            color=color,\n",
    "                            alpha=0.5)\n",
    "                ax_clsd.fill_betweenx(\n",
    "                            epochs,\n",
    "                            [float(x_values)]*len(probs),\n",
    "                            float(x_values)*2**(np.std(y_values)*probs),\n",
    "                            color=color,\n",
    "                            alpha=0.1)\n",
    "            if y_mean_marker is not None:\n",
    "                ax_clsd.scatter([x_values],\n",
    "                         [np.mean(y_values)],\n",
    "                         s=25,\n",
    "                         color=color,\n",
    "                         alpha=1.,\n",
    "                         marker=y_mean_marker, \n",
    "                         facecolors='white')\n",
    "        elif submission_stat['division'] == 'open':\n",
    "            y_values_unique = np.unique(y_values)\n",
    "            y_values_frequency = np.array([np.sum(e == np.array(y_values)) for e in y_values_unique])\n",
    "            ax_open.scatter([x_values]*len(y_values_unique),\n",
    "                        y_values_unique,\n",
    "                        s=5*y_values_frequency,\n",
    "                        color=color,\n",
    "                        label=r'$\\bigstar$' + submission_stat['submission_label'] + (f' ({color_label(submission_stat)})' if color_label is not None else ''),\n",
    "                        alpha=1.)\n",
    "            ax_open.annotate(r'$\\bigstar$' + submission_annotation,\n",
    "                        xy=(x_values, np.mean(y_values)),\n",
    "                        color=color,\n",
    "                        **submission_annotation_opts)\n",
    "            if secondary_x_key is not None:\n",
    "                ax_open.plot([secondary_x_values, x_values], # requires secondary_x to primary x scale conversion if non-identity\n",
    "                         [np.mean(y_values)]*2,\n",
    "                         '--',\n",
    "                         color=color,\n",
    "                         alpha=0.5)\n",
    "                ax_open.scatter([secondary_x_values], # requires secondary_x to primary x scale conversion if non-identity\n",
    "                         [np.mean(y_values)],\n",
    "                         s=25,\n",
    "                         color=color,\n",
    "                         alpha=1.,\n",
    "                         marker=secondary_x_marker, \n",
    "                         facecolors='white')\n",
    "            if secondary_y_keys is not None:\n",
    "                for i in np.nonzero(secondary_y_keys_not_none)[0]: \n",
    "                    x_factor = 2**(0.05*(i+1) if benchmark.name != 'cosmoflow' else 0.1)\n",
    "                    ax_clsd.errorbar([x_values/x_factor],\n",
    "                                     [np.mean(secondary_y_values[i])],\n",
    "                                yerr=[np.std(secondary_y_values[i])],\n",
    "                                fmt=secondary_y_markers[i], \n",
    "                                color=color,\n",
    "                                alpha=1.)\n",
    "                    circle_y, circle_x = horizontal_circle_segment_loglog([np.mean(y_values), np.mean(secondary_y_values[i])], x_values, y_factor=x_factor)\n",
    "                    ax_clsd.plot(circle_x,\n",
    "                             circle_y,\n",
    "                             color=color,\n",
    "                             alpha=.3)\n",
    "            if len(y_values_unique) != 1:\n",
    "                epochs, probs = kde_estimate(y_values)\n",
    "                ax_open.plot(   float(x_values)*2**(np.std(y_values)*probs),\n",
    "                            epochs,\n",
    "                            color=color,\n",
    "                            alpha=0.5)\n",
    "                ax_open.fill_betweenx(\n",
    "                            epochs,\n",
    "                            [float(x_values)]*len(probs),\n",
    "                            float(x_values)*2**(np.std(y_values)*probs),\n",
    "                            color=color,\n",
    "                            alpha=0.1)\n",
    "            if y_mean_marker is not None:\n",
    "                ax_open.scatter([x_values],\n",
    "                         [np.mean(y_values)],\n",
    "                         s=25,\n",
    "                         color=color,\n",
    "                         alpha=1.,\n",
    "                         marker=y_mean_marker, \n",
    "                         facecolors='white')\n",
    "        else:\n",
    "            raise ValueError(f\"Invalid division {submission_stat['division']}.\")\n",
    "\n",
    "    if legend_side is not None:\n",
    "        ax_clsd.legend(bbox_to_anchor=( 1.1 if ax_open == ax_clsd and legend_side == 'right' else -0.8, 1), loc='upper left', frameon=False, fontsize='small')\n",
    "        if ax_open != ax_clsd:\n",
    "            ax_open.legend(bbox_to_anchor=(1.55, 1), loc='upper right', frameon=False, fontsize='small')\n",
    "\n",
    "    if show_symbolic_legend:\n",
    "\n",
    "        symbolic_legend_labels = []\n",
    "        ax_clsd_twin = ax_clsd.twinx()\n",
    "        \n",
    "        # Symbolic legend\n",
    "        ax_clsd_twin.scatter([], [], label=\"Symbolic legend\", marker=\"None\") \n",
    "        symbolic_legend_labels.append(\"Symbolic legend\")\n",
    "        ax_clsd_twin.scatter([], [], label=y_legend_label, color='black')\n",
    "        symbolic_legend_labels.append(y_legend_label)\n",
    "        if secondary_x_key is not None:\n",
    "            ax_clsd_twin.plot([], [], secondary_x_marker, fillstyle='none', label=secondary_x_legend_label, color='black')\n",
    "            symbolic_legend_labels.append(secondary_x_legend_label)\n",
    "        if secondary_y_keys is not None:\n",
    "            for i,_ in enumerate(secondary_y_keys):\n",
    "                ax_clsd_twin.plot([], [], secondary_y_markers[i], label=secondary_y_legend_labels[i], color='black')\n",
    "                symbolic_legend_labels.append(secondary_y_legend_labels[i])\n",
    "\n",
    "        # Color-coding\n",
    "        ax_clsd_twin.scatter([], [], label=\" \", marker=\"None\") \n",
    "        symbolic_legend_labels.append(\" \")\n",
    "        ax_clsd_twin.scatter([], [], label=\"Color-coding\", marker=\"None\") \n",
    "        symbolic_legend_labels.append(\"Color-coding\")\n",
    "        for color_key, color in colors.items():\n",
    "            ax_clsd_twin.scatter([], [], label=color_legend_label[color_key] if color_legend_label is not None else color_key, color=color, marker='o')\n",
    "            symbolic_legend_labels.append(color_legend_label[color_key] if color_legend_label is not None else color_key)\n",
    "        \n",
    "        ax_clsd_twin.get_yaxis().set_visible(False)\n",
    "\n",
    "        legend_handles, legend_labels = ax_clsd_twin.get_legend_handles_labels()\n",
    "        legend_labels, legend_handles = zip(*sorted(zip(legend_labels, legend_handles), key=lambda t: symbolic_legend_labels.index(t[0])))\n",
    "        ax_clsd_twin.legend(legend_handles, legend_labels, bbox_to_anchor=( 1.5 if legend_side == 'right' else -0.15, 0), loc='lower right', frameon=False, fontsize='small')\n",
    "\n",
    "    \n",
    "    ax_clsd.grid(True, which='major', alpha=0.3)\n",
    "    ax_open.grid(True, which='major', alpha=0.3)\n",
    "\n",
    "    if axis_scale == 'linear':\n",
    "        ax_clsd.plot()\n",
    "        ax_open.plot()\n",
    "    elif axis_scale == 'semilogx':\n",
    "        ax_clsd.semilogx()\n",
    "        ax_open.semilogx()\n",
    "    elif axis_scale == 'loglog':\n",
    "        ax_clsd.loglog()\n",
    "        ax_open.loglog()\n",
    "    else:\n",
    "        raise ValueError(f'axis_scale should be in [semilogx, loglog, plot], not {axis_scale}')\n",
    "\n",
    "    if ax_clsd != ax_open:\n",
    "        if math.isfinite(x_values_min['closed']) and math.isfinite(x_values_max['closed']):\n",
    "            ax_clsd.set_xlim(left=x_values_min['closed']//2 if axis_scale != 'linear' else 0., right=x_values_max['closed']*2)\n",
    "        if math.isfinite(x_values_min['open']) and math.isfinite(x_values_max['open']):\n",
    "            ax_open.set_xlim(left=x_values_min['open']//2 if axis_scale != 'linear' else 0.,   right=x_values_max['open']*2)\n",
    "    else:\n",
    "        if math.isfinite(x_values_min['closed']) and math.isfinite(x_values_max['closed']):\n",
    "            ax_clsd.set_xlim(left=min(x_values_min['closed'], x_values_min['open'])//2 if axis_scale != 'linear' else 0., right=max(x_values_max['closed'], x_values_max['open'])*2)\n",
    "            \n",
    "    if revert_x_axis:\n",
    "        xlim = ax_clsd.get_xlim()\n",
    "        ax_clsd.set_xlim(xlim[1], xlim[0])\n",
    "\n",
    "    if revert_y_axis:\n",
    "        ylim = ax_clsd.get_ylim()\n",
    "        ax_clsd.set_ylim(ylim[1], ylim[0])\n",
    "\n",
    "    ax_clsd.xaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "    ax_open.xaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "\n",
    "    ax_clsd.xaxis.set_major_formatter(plt.FormatStrFormatter('%d'))\n",
    "    ax_open.xaxis.set_major_formatter(plt.ScalarFormatter())\n",
    "\n",
    "    if format_y_axis_as_x_axis:\n",
    "        ax_clsd.yaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "        ax_open.yaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "\n",
    "        ax_clsd.yaxis.set_major_formatter(plt.FormatStrFormatter('%d'))\n",
    "        ax_open.yaxis.set_major_formatter(plt.ScalarFormatter())\n",
    "\n",
    "    if secondary_y_keys is not None:\n",
    "        secondary_y_ax_clsd.set_ylabel(secondary_y_axis_label if secondary_y_axis_label is not None else secondary_y_keys[0])\n",
    "        secondary_y_ax_open.set_ylabel(secondary_y_axis_label if secondary_y_axis_label is not None else secondary_y_keys[0])\n",
    "\n",
    "    \n",
    "    if secondary_x_key is not None: # does the following have an effect at all?        \n",
    "        if secondary_x_ticks_same_as_primary:\n",
    "            secondary_x_ax_clsd.tick_params(axis='x', top=False, labeltop=False)\n",
    "            secondary_x_ax_open.tick_params(axis='x', top=False, labeltop=False)\n",
    "            ax_clsd.tick_params(axis='x', top=True, labeltop=True)\n",
    "            ax_open.tick_params(axis='x', top=True, labeltop=True)\n",
    "        else:\n",
    "            secondary_x_ax_clsd.xaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "            secondary_x_ax_open.xaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "\n",
    "            secondary_x_ax_clsd.xaxis.set_major_formatter(plt.FormatStrFormatter('%d'))\n",
    "        secondary_x_ax_clsd.set_xlabel(secondary_x_label if secondary_x_label is not None else secondary_x_key)\n",
    "        secondary_x_ax_open.set_xlabel(secondary_x_label if secondary_x_label is not None else secondary_x_key)\n",
    "    \n",
    "    if level_func is not None:\n",
    "        if level_transform is not None:\n",
    "            if len(inspect.signature(level_transform[0]).parameters) > 1: # submission-dependent transformation\n",
    "                level_transform_partial = (functools.partial(level_transform[0], s=submission_stat),\n",
    "                                           functools.partial(level_transform[1], s=submission_stat))\n",
    "            else:\n",
    "                level_transform_partial = level_transform\n",
    "\n",
    "        plot_level_lines(ax=ax_clsd, \n",
    "                        x_range=(min(x_values_min['closed'],x_values_min['open']), max(x_values_max['closed'],x_values_max['open'])), \n",
    "                        y_range=(min(y_values_min['closed'],y_values_min['open']), max(y_values_max['closed'],y_values_max['open'])), \n",
    "                        level_func=level_func, \n",
    "                        level_label=level_label,\n",
    "                        level_transform=level_transform_partial if level_transform is not None else None,\n",
    "                        level_color=level_color, \n",
    "                        level_log_base=level_log_base,\n",
    "                        level_fine_grid=level_fine_grid)\n",
    "        \n",
    "    ax_clsd.set_xlabel(x_label if x_label is not None else x_key)\n",
    "    ax_open.set_xlabel(x_label if x_label is not None else x_key)\n",
    "    ax_clsd.set_ylabel(y_label)\n",
    "    ax_open.set_ylabel(y_label)\n",
    "\n",
    "    if secondary_y_keys is not None:\n",
    "        ax_clsd.yaxis.tick_right()\n",
    "        ax_open.yaxis.tick_right()\n",
    "        ax_clsd.yaxis.set_label_position('right')\n",
    "        ax_open.yaxis.set_label_position('right')\n",
    "    \n",
    "    if fig is not None:\n",
    "#         fig.suptitle(benchmark.round + '/' + benchmark.name + ' - '+ (title if title is not None else f'{x_label} vs. {y_label}'))\n",
    "        fig.tight_layout()\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "def horizontal_circle_segment_loglog(x_values, y_value, y_factor=2**0.1):\n",
    "    assert y_factor > 1\n",
    "    log_x_values = np.log(x_values)\n",
    "    log_y_values = np.log([y_value, y_value/y_factor])\n",
    "    \n",
    "    log_p0 = np.array([log_x_values[0], log_y_values[0]])\n",
    "    log_p1 = np.array([log_x_values[1], log_y_values[1]])\n",
    " \n",
    "    log_center_x = 0.5 * np.linalg.norm(log_p1-log_p0)\n",
    "    log_x_unit = (log_p1-log_p0)/np.linalg.norm(log_p1-log_p0)\n",
    "    log_x_unit_angle = np.arctan(log_x_unit[1]/log_x_unit[0]) if log_x_unit[0] != 0 else (np.pi/2. if log_x_unit[1] > 0 else -np.pi/2.)\n",
    "    \n",
    "    log_alpha = np.arctan( min(1, np.log(y_factor)/log_center_x) )\n",
    "    log_r = log_center_x/np.sin(log_alpha)\n",
    "    log_center_y = np.sign(log_x_unit[0])*log_center_x/np.tan(log_alpha)\n",
    "\n",
    "    log_center = np.dot(np.array([[log_x_unit[0], -log_x_unit[1]], [log_x_unit[1], log_x_unit[0]]]), np.array([log_center_x, log_center_y]).transpose()) + log_p0\n",
    "        \n",
    "    log_angles = np.linspace(-log_alpha+log_x_unit_angle, log_alpha+log_x_unit_angle, 25)\n",
    "    return np.exp(log_center[0] + log_r*np.sin(log_angles)), np.exp(log_center[1] - log_r*np.cos(log_angles))\n",
    "\n",
    "# #Testing\n",
    "# x_values = [0.5, 0.49]\n",
    "# y_value = 1\n",
    "# y_factor = 2**0.01\n",
    "# circle_x, circle_y = horizontal_circle_segment_loglog(x_values, y_value, y_factor)\n",
    "# print(circle_x)\n",
    "# print(circle_y)\n",
    "# print(np.exp(np.log(2)))\n",
    "# plt.loglog(circle_x, circle_y)\n",
    "# plt.loglog(x_values, [y_value]*2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Transposition of roles of x and y in plot above\n",
    "\n",
    "def plot_all_submissions_flipped(y_key, x_key, y_label=None, x_label=None, x_transform=None, x_legend_label=None,\n",
    "                                 secondary_x_keys=None, secondary_x_axis_label=None, secondary_x_legend_labels=None, secondary_x_transforms=None, secondary_x_markers=None, primary_to_secondary_x_transform=None, secondary_to_primary_x_transform=None, \n",
    "                                 secondary_y_key=None, secondary_y_label=None, secondary_y_legend_label=None, secondary_y_marker=None, primary_to_secondary_y_transform=None, secondary_to_primary_y_transform=None, \n",
    "                                 level_func=None, level_label=None, level_transform=None, level_color='black', level_log_base=10., level_fine_grid=1, format_x_axis_as_y_axis=False,\n",
    "                                 title=None, sharey=False, axis_scale='semilogx', filter_submissions=lambda s: True, color_label=None,  color_label_range=None, show_symbolic_legend=False, \n",
    "                                 submission_label_annotation_opts=None, submission_annotation_key='submission_short_label', ax=None):\n",
    "    \n",
    "    if type(y_key) is str:\n",
    "        assert not y_key.startswith('run_')\n",
    "    if type(secondary_y_key) is str:\n",
    "        assert not secondary_y_key.startswith('run_')\n",
    "    if type(x_key) is str:\n",
    "        assert x_key.startswith('run_')\n",
    "    if secondary_x_keys is not None:\n",
    "        for secondary_x_key in secondary_x_keys:\n",
    "            if type(secondary_x_key) is str:\n",
    "                assert secondary_x_key.startswith('run_')\n",
    "\n",
    "    fig = None\n",
    "    if ax is None:\n",
    "        fig, ax = plt.subplots(ncols=1,\n",
    "                               figsize=(24, 12))\n",
    "    ax_clsd = ax\n",
    "    ax_open = ax\n",
    "    \n",
    "    colors = create_submission_colors(color_label=color_label, color_label_range=color_label_range, filter_submissions=filter_submissions)    \n",
    "\n",
    "    if secondary_x_keys is not None:\n",
    "        secondary_ax_clsd = ax_clsd.secondary_xaxis('top', functions=(primary_to_secondary_x_transform, secondary_to_primary_x_transform)) # primary x to secondary x conversion and vice versa tuple for non-identity\n",
    "        if ax_open != ax_clsd:\n",
    "            secondary_ax_open = ax_open.secondary_xaxis('top', functions=(primary_to_secondary_x_transform, secondary_to_primary_x_transform))\n",
    "        else:\n",
    "            secondary_ax_open = secondary_ax_clsd\n",
    "\n",
    "    if secondary_y_key is not None:\n",
    "        secondary_y_ax_clsd = ax_clsd.secondary_yaxis('right', functions=(primary_to_secondary_y_transform, secondary_to_primary_y_transform)) # primary x to secondary x conversion and vice versa tuple for non-identity\n",
    "        if ax_open != ax_clsd:\n",
    "            secondary_y_ax_open = ax_open.secondary_yaxis('right', functions=(primary_to_secondary_y_transform, secondary_to_primary_y_transform))\n",
    "        else:\n",
    "            secondary_y_ax_open = secondary_y_ax_clsd\n",
    "\n",
    "    x_values_min = dict(closed=float('inf'), open=float('inf')) \n",
    "    x_values_max = dict(closed=float('-inf'), open=float('-inf'))\n",
    "    y_values_min = dict(closed=float('inf'), open=float('inf')) \n",
    "    y_values_max = dict(closed=float('-inf'), open=float('-inf'))\n",
    "    \n",
    "    # Plot all submissions\n",
    "    for i, (k, submission_stat) in enumerate(submissions_stats.items()):\n",
    "        \n",
    "        if not filter_submissions(submission_stat):\n",
    "            continue\n",
    "\n",
    "        submission_annotation = submission_stat[submission_annotation_key] if type(submission_annotation_key) is str else submission_annotation_key(submission_stat)\n",
    "        if submission_label_annotation_opts is None:\n",
    "            submission_annotation_opts = dict(\n",
    "                        xytext=( 0, -8),\n",
    "                        textcoords=\"offset points\", \n",
    "                        ha='right', va='top', rotation=90, #rotation=49,            \n",
    "                        alpha=1., fontsize=12)\n",
    "        else:\n",
    "            submission_annotation_opts = submission_label_annotation_opts if type(submission_label_annotation_opts) is dict else submission_label_annotation_opts(submission_stat)\n",
    "\n",
    "        color = colors[k if color_label is None else color_label(submission_stat)]\n",
    "            \n",
    "        y_values = submission_stat[y_key] if type(y_key) is str else y_key(submission_stat)\n",
    "        y_values_min[submission_stat['division']] = min(y_values_min[submission_stat['division']], np.min(y_values))\n",
    "        y_values_max[submission_stat['division']] = max(y_values_max[submission_stat['division']], np.max(y_values))\n",
    "        if secondary_y_key is not None:\n",
    "            secondary_y_values = submission_stat[secondary_y_key] if type(secondary_y_key) is str else secondary_y_key(submission_stat)\n",
    "            secondary_y_values = secondary_to_primary_y_transform(secondary_y_values)\n",
    "            y_values_min[submission_stat['division']] = min(y_values_min[submission_stat['division']], np.min(secondary_y_values))\n",
    "            y_values_max[submission_stat['division']] = max(y_values_max[submission_stat['division']], np.max(secondary_y_values))\n",
    "\n",
    "        submission_stat_runs = sorted(submission_stat['runs'], key=lambda run: run['run_time'])\n",
    "        submission_stat_runs = submission_stat_runs[1:mlperf_opts[benchmark.round][benchmark.name]['runs_to_submit']-1] # olympic scoring rules\n",
    "\n",
    "        x_values = np.array([run[x_key] if type(x_key) is str else x_key({'run': run, 'system': submission_stat['system']}) for run in submission_stat_runs])\n",
    "        if secondary_x_keys is not None:\n",
    "            secondary_x_keys_not_none = []\n",
    "            secondary_x_values = []\n",
    "            for secondary_x_key in secondary_x_keys:\n",
    "                secondary_x_values.append( np.array([run[secondary_x_key] if type(secondary_x_key) is str else secondary_x_key({'run': run, 'system': submission_stat['system']}) for run in submission_stat_runs]) )\n",
    "                secondary_x_keys_not_none.append(None not in secondary_x_values[-1])\n",
    "                if secondary_x_keys_not_none[-1]:\n",
    "                    secondary_x_values[-1] = np.array([secondary_to_primary_x_transform(val) for val in secondary_x_values[-1]])\n",
    "        \n",
    "        if x_transform is not None:\n",
    "            x_values = np.array([x_transform(val) for val in x_values])\n",
    "        x_values_min[submission_stat['division']] = min(x_values_min[submission_stat['division']], np.min(x_values))\n",
    "        x_values_max[submission_stat['division']] = max(x_values_max[submission_stat['division']], np.max(x_values))\n",
    "\n",
    "        if secondary_x_keys is not None:\n",
    "            for i in np.nonzero(secondary_x_keys_not_none)[0]: \n",
    "                if secondary_x_transforms[i] is not None:\n",
    "                    secondary_x_values[i] = np.array([secondary_x_transforms[i](val) for val in secondary_x_values[i]])\n",
    "                x_values_min[submission_stat['division']] = min(x_values_min[submission_stat['division']], np.min(secondary_x_values[i]))\n",
    "                x_values_max[submission_stat['division']] = max(x_values_max[submission_stat['division']], np.max(secondary_x_values[i]))\n",
    "        \n",
    "\n",
    "        if submission_stat['division'] == 'closed':                \n",
    "            x_values_unique = np.unique(x_values)\n",
    "            x_values_frequency = np.array([np.sum(e == np.array(x_values)) for e in x_values_unique])\n",
    "            ax_clsd.scatter(x_values_unique,\n",
    "                        [y_values]*len(x_values_unique),\n",
    "                        s=5*x_values_frequency,\n",
    "                        color=color,\n",
    "                        label=submission_stat['submission_label'] + (f' ({color_label(submission_stat)})' if color_label is not None else ''),\n",
    "                        alpha=1.)\n",
    "            ax_clsd.annotate(submission_annotation,\n",
    "                        xy=(np.mean(x_values), y_values),\n",
    "                        color=color,\n",
    "                        **submission_annotation_opts)\n",
    "            if secondary_y_key is not None:                \n",
    "                ax_clsd.plot([np.mean(x_values)]*2, \n",
    "                        [secondary_y_values, y_values], # requires secondary_x to primary x scale conversion if non-identity\n",
    "                         '--',\n",
    "                         color=color,\n",
    "                         alpha=0.3)\n",
    "                ax_clsd.scatter([np.mean(x_values)],\n",
    "                         [secondary_y_values], # requires secondary_x to primary x scale conversion if non-identity\n",
    "                         s=25,\n",
    "                         color=color,\n",
    "                         alpha=1.,\n",
    "                         marker=secondary_y_marker)                                           \n",
    "            if secondary_x_keys is not None:\n",
    "                for i in np.nonzero(secondary_x_keys_not_none)[0]: #for i,_ in enumerate(secondary_x_keys):\n",
    "                    y_factor = 2**(0.05*(i+1) if benchmark.name != 'cosmoflow' else 0.1)\n",
    "                    ax_clsd.errorbar([np.mean(secondary_x_values[i])],\n",
    "                                [y_values/y_factor],\n",
    "                                xerr=[np.std(secondary_x_values[i])],\n",
    "                                fmt=secondary_x_markers[i], #markersize=3,\n",
    "                                color=color,\n",
    "                                alpha=1.)\n",
    "                    circle_x, circle_y = horizontal_circle_segment_loglog([np.mean(x_values), np.mean(secondary_x_values[i])], y_values, y_factor=y_factor)\n",
    "                    ax_clsd.plot(circle_x,\n",
    "                             circle_y,\n",
    "                             color=color,\n",
    "                             alpha=.3)\n",
    "            if len(x_values_unique) != 1:\n",
    "                epochs, probs = kde_estimate(x_values)\n",
    "                ax_clsd.plot(   epochs,\n",
    "                            float(y_values)*2**(np.std(x_values)*probs),\n",
    "                            color=color,\n",
    "                            alpha=0.5)\n",
    "                ax_clsd.fill_between(\n",
    "                            epochs,\n",
    "                            [float(y_values)]*len(probs),\n",
    "                            float(y_values)*2**(np.std(x_values)*probs),\n",
    "                            color=color,\n",
    "                            alpha=0.1)\n",
    "        elif submission_stat['division'] == 'open':\n",
    "            x_values_unique = np.unique(x_values)\n",
    "            x_values_frequency = np.array([np.sum(e == np.array(x_values)) for e in x_values_unique])\n",
    "            ax_open.scatter(x_values_unique,\n",
    "                        [y_values]*len(x_values_unique),\n",
    "                        s=5*x_values_frequency,\n",
    "                        color=color,\n",
    "                        label=r'$\\bigstar$' + submission_stat['submission_label'] + (f' ({color_label(submission_stat)})' if color_label is not None else ''),\n",
    "                        alpha=1.)\n",
    "            ax_open.annotate(r'$\\bigstar$' + submission_annotation,\n",
    "                        xy=(np.mean(x_values), y_values),\n",
    "                        color=color,\n",
    "                        **submission_annotation_opts)\n",
    "            if secondary_y_key is not None:\n",
    "                ax_open.plot([np.mean(x_values)]*2, \n",
    "                        [secondary_y_values, y_values], # requires secondary_x to primary x scale conversion if non-identity\n",
    "                         '--',\n",
    "                         color=color,\n",
    "                         alpha=0.3)\n",
    "                ax_open.scatter([np.mean(x_values)],\n",
    "                         [secondary_y_values], # requires secondary_x to primary x scale conversion if non-identity\n",
    "                         s=25,\n",
    "                         color=color,\n",
    "                         alpha=1.,\n",
    "                         marker=secondary_y_marker)                                           \n",
    "            if secondary_x_keys is not None:\n",
    "                for i in np.nonzero(secondary_x_keys_not_none)[0]: #for i,_ in enumerate(secondary_x_keys):\n",
    "                    y_factor = 2**(0.05*(i+1) if benchmark.name != 'cosmoflow' else 0.1)\n",
    "                    ax_open.errorbar([np.mean(secondary_x_values[i])],\n",
    "                                [y_values/y_factor],\n",
    "                                xerr=[np.std(secondary_x_values[i])],\n",
    "                                fmt=secondary_x_markers[i], #markersize=3,\n",
    "                                color=color,\n",
    "                                alpha=1.)\n",
    "                    circle_x, circle_y = horizontal_circle_segment_loglog([np.mean(x_values), np.mean(secondary_x_values[i])], y_values, y_factor=y_factor)\n",
    "                    ax_open.plot(circle_x,\n",
    "                             circle_y,\n",
    "                             color=color,\n",
    "                             alpha=0.3)\n",
    "            if len(x_values_unique) != 1:\n",
    "                epochs, probs = kde_estimate(x_values)\n",
    "                ax_open.plot(   epochs,\n",
    "                            float(y_values)*2**(np.std(x_values)*probs),\n",
    "                            color=color,\n",
    "                            alpha=0.5)\n",
    "                ax_open.fill_between(\n",
    "                            epochs,\n",
    "                            [float(y_values)]*len(probs),\n",
    "                            float(y_values)*2**(np.std(x_values)*probs),\n",
    "                            color=color,\n",
    "                            alpha=0.1)\n",
    "        else:\n",
    "            raise ValueError(f\"Invalid division {submission_stat['division']}.\")\n",
    "    \n",
    "    ax_clsd.legend(bbox_to_anchor=( 1.1, 1), loc='upper left', frameon=False, fontsize='small')\n",
    "    if ax_open != ax_clsd:\n",
    "        ax_open.legend(bbox_to_anchor=(1.55, 1), loc='upper right', frameon=False, fontsize='small')\n",
    "\n",
    "    if show_symbolic_legend:\n",
    "        symbolic_legend_labels = []\n",
    "        ax_clsd_twin = ax_clsd.twinx()\n",
    "        \n",
    "        # Symbolic legend\n",
    "        ax_clsd_twin.scatter([], [], label=\"Symbolic legend\", marker=\"None\") \n",
    "        symbolic_legend_labels.append(\"Symbolic legend\")\n",
    "        ax_clsd_twin.scatter([], [], label=x_legend_label, color='black')\n",
    "        symbolic_legend_labels.append(x_legend_label)\n",
    "        if secondary_y_key is not None:\n",
    "            ax_clsd_twin.plot([], [], secondary_y_marker, label=secondary_y_legend_label, color='black')\n",
    "            symbolic_legend_labels.append(secondary_y_legend_label)\n",
    "        if secondary_x_keys is not None:\n",
    "            for i,_ in enumerate(secondary_x_keys):\n",
    "                ax_clsd_twin.plot([], [], secondary_x_markers[i], label=secondary_x_legend_labels[i], color='black')\n",
    "                symbolic_legend_labels.append(secondary_x_legend_labels[i])\n",
    "\n",
    "        # Color-coding\n",
    "        ax_clsd_twin.scatter([], [], label=\" \", marker=\"None\") \n",
    "        symbolic_legend_labels.append(\" \")\n",
    "        ax_clsd_twin.scatter([], [], label=\"Color-coding\", marker=\"None\") \n",
    "        symbolic_legend_labels.append(\"Color-coding\")\n",
    "        for color_key, color in colors.items():\n",
    "            ax_clsd_twin.scatter([], [], label=color_key, color=color, marker='o')\n",
    "            symbolic_legend_labels.append(color_key)\n",
    "\n",
    "        ax_clsd_twin.get_yaxis().set_visible(False)\n",
    "\n",
    "        legend_handles, legend_labels = ax_clsd_twin.get_legend_handles_labels()\n",
    "        legend_labels, legend_handles = zip(*sorted(zip(legend_labels, legend_handles), key=lambda t: symbolic_legend_labels.index(t[0])))\n",
    "        ax_clsd_twin.legend(legend_handles, legend_labels, bbox_to_anchor=( 1.6, 0), loc='lower right', frameon=False, fontsize='small')\n",
    "    \n",
    "    ax_clsd.grid(True, which='major', alpha=0.3)\n",
    "    ax_open.grid(True, which='major', alpha=0.3)\n",
    "\n",
    "    if axis_scale == 'linear':\n",
    "        ax_clsd.plot()\n",
    "        ax_open.plot()\n",
    "    elif axis_scale == 'semilogx':\n",
    "        ax_clsd.semilogx()\n",
    "        ax_open.semilogx()\n",
    "    elif axis_scale == 'loglog':\n",
    "        ax_clsd.loglog()\n",
    "        ax_open.loglog()\n",
    "    else:\n",
    "        raise ValueError(f'axis_scale should be in [semilogx, loglog, plot], not {axis_scale}')\n",
    "        \n",
    "\n",
    "          \n",
    "    if level_func is not None:\n",
    "        if level_transform is not None:\n",
    "            if len(inspect.signature(level_transform[0]).parameters) > 1: # submission-dependent transformation\n",
    "                level_transform_partial = (functools.partial(level_transform[0], s=submission_stat),\n",
    "                                           functools.partial(level_transform[1], s=submission_stat))\n",
    "            else:\n",
    "                level_transform_partial = level_transform\n",
    "\n",
    "        plot_level_lines(ax=ax_clsd, \n",
    "                        x_range=(min(x_values_min['closed'],x_values_min['open']), max(x_values_max['closed'],x_values_max['open'])), \n",
    "                        y_range=(min(y_values_min['closed'],y_values_min['open']), max(y_values_max['closed'],y_values_max['open'])), \n",
    "                        level_func=level_func, \n",
    "                        level_label=level_label,\n",
    "                        level_transform=level_transform_partial if level_transform is not None else None,\n",
    "                        level_color=level_color, \n",
    "                        level_log_base=level_log_base,\n",
    "                        level_fine_grid=level_fine_grid)\n",
    "                                         \n",
    "    if ax_clsd != ax_open:\n",
    "        if math.isfinite(y_values_min['closed']) and math.isfinite(y_values_max['closed']):\n",
    "            ax_clsd.set_ylim(bottom=y_values_min['closed']//2 if axis_scale != 'linear' else 0., top=y_values_max['closed']*2)\n",
    "        if math.isfinite(y_values_min['open']) and math.isfinite(y_values_max['open']):\n",
    "            ax_open.set_ylim(bottom=y_values_min['open']//2 if axis_scale != 'linear' else 0.,   top=y_values_max['open']*2)\n",
    "    else:\n",
    "        if math.isfinite(y_values_min['closed']) and math.isfinite(y_values_max['closed']):\n",
    "            ax_clsd.set_ylim(bottom=min(y_values_min['closed']//2, y_values_min['open']//2) if axis_scale != 'linear' else 0., top=max(y_values_max['closed']*2, y_values_max['open']*2))\n",
    "        \n",
    "\n",
    "\n",
    "    ax_clsd.yaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "    ax_open.yaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "\n",
    "    ax_clsd.yaxis.set_major_formatter(plt.FormatStrFormatter('%d'))\n",
    "    ax_open.yaxis.set_major_formatter(plt.ScalarFormatter())\n",
    "    \n",
    "    if format_x_axis_as_y_axis:\n",
    "        ax_clsd.xaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "        ax_open.xaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "\n",
    "        ax_clsd.xaxis.set_major_formatter(plt.FormatStrFormatter('%d'))\n",
    "        ax_open.xaxis.set_major_formatter(plt.ScalarFormatter())\n",
    "\n",
    "    if secondary_x_keys is not None:\n",
    "        secondary_ax_clsd.xaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "        secondary_ax_open.xaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "\n",
    "        secondary_ax_clsd.xaxis.set_major_formatter(plt.FormatStrFormatter('%d'))\n",
    "        secondary_ax_open.xaxis.set_major_formatter(plt.ScalarFormatter())\n",
    "        secondary_ax_clsd.set_xlabel(secondary_x_axis_label if secondary_x_axis_label is not None else secondary_x_keys[0])\n",
    "        secondary_ax_open.set_xlabel(secondary_x_axis_label if secondary_x_axis_label is not None else secondary_x_keys[0])\n",
    "\n",
    "    if secondary_y_key is not None:\n",
    "        secondary_y_ax_clsd.yaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "        secondary_y_ax_open.yaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "\n",
    "        secondary_y_ax_clsd.yaxis.set_major_formatter(plt.FormatStrFormatter('%d'))\n",
    "        secondary_y_ax_open.yaxis.set_major_formatter(plt.ScalarFormatter())\n",
    "        secondary_y_ax_clsd.set_ylabel(secondary_y_label if secondary_y_label is not None else secondary_y_key)\n",
    "        secondary_y_ax_open.set_ylabel(secondary_y_label if secondary_y_label is not None else secondary_y_key)\n",
    "    \n",
    "    ax_clsd.set_ylabel(y_label if y_label is not None else y_key)\n",
    "    ax_open.set_ylabel(y_label if y_label is not None else y_key)\n",
    "    ax_clsd.set_xlabel(x_label)\n",
    "    ax_open.set_xlabel(x_label)\n",
    "\n",
    "    if fig is not None:\n",
    "#         fig.suptitle(benchmark.round + '/' + benchmark.name + ' - '+ (title if title is not None else f'{y_label} vs. {x_label}'))\n",
    "        fig.tight_layout()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_directed_x_marker(value, reference):\n",
    "    if value < reference:\n",
    "        return '<'\n",
    "    elif value > reference:\n",
    "        return '>'\n",
    "    else:\n",
    "        return 's'\n",
    "\n",
    "def get_directed_y_marker(value, reference):\n",
    "    if value < reference:\n",
    "        return 'v'\n",
    "    elif value > reference:\n",
    "        return '^'\n",
    "    else:\n",
    "        return 's'\n",
    "\n",
    "def plot_all_submissions_scatter_vector(x_key, y_key, x_label=None, y_label=None, x_legend_label=None, x_transform=None, y_transform=None, \n",
    "                                        vector_x_key=None, vector_x_label=None,  vector_x_legend_label=None, primary_to_vector_x_transform=None, vector_to_primary_x_transform=None,\n",
    "                                        vector_y_key=None, vector_y_label=None,  vector_y_legend_label=None, primary_to_vector_y_transform=None, vector_to_primary_y_transform=None,\n",
    "                                        vector_annotate_label=None,\n",
    "                                        xy_mean_marker=None,\n",
    "                                        title=None, sharey=False, axis_scale='semilogx', \n",
    "                                        filter_submissions=lambda s: True, color_label=None,  color_label_range=None, ax=None,\n",
    "                                        level_func=None, level_label=None, level_transform=None, level_color='black', level_log_base=10., level_fine_grid=1,\n",
    "                                        vector_level_func=None, vector_level_label=None, vector_level_transform=None, vector_level_color='red', vector_level_log_base=2.,\n",
    "                                        submission_annotation_key='submission_short_label', submission_label_annotation_opts=None):\n",
    "    \n",
    "    if type(x_key) is str:\n",
    "        assert x_key.startswith('run_')\n",
    "    if type(y_key) is str:\n",
    "        assert y_key.startswith('run_')\n",
    "\n",
    "    fig = None\n",
    "    if ax is None:\n",
    "        fig, ax = plt.subplots(ncols=1, sharey=sharey, \n",
    "                            figsize=(24, 12))\n",
    "    ax_clsd = ax\n",
    "    ax_open = ax\n",
    "\n",
    "    if submission_label_annotation_opts is None:\n",
    "        submission_label_annotation_opts = dict(\n",
    "                        xytext=(-10, 0),\n",
    "                        textcoords=\"offset points\", \n",
    "                        ha='right', va='top',  rotation=45 if vector_x_key is not None or vector_y_key is not None else 0,\n",
    "                        alpha=1., fontsize=12)\n",
    "    colors = create_submission_colors(color_label=color_label, color_label_range=color_label_range, filter_submissions=filter_submissions)\n",
    "\n",
    "\n",
    "    if vector_x_key is not None:\n",
    "        assert (primary_to_vector_x_transform is not None) == (vector_to_primary_x_transform is not None)\n",
    "        vector_x_axis_functions = (primary_to_vector_x_transform, vector_to_primary_x_transform) if (primary_to_vector_x_transform is not None) and (vector_to_primary_x_transform is not None) else None\n",
    "        vector_x_ax_clsd = ax_clsd.secondary_xaxis('top', functions=vector_x_axis_functions) # FIXME: primary x to vector x conversion and vice versa tuple for non-identity\n",
    "        if ax_open != ax_clsd:\n",
    "            vector_x_ax_open = ax_open.secondary_xaxis('top', functions=vector_x_axis_functions)\n",
    "        else:\n",
    "            vector_x_ax_open = vector_x_ax_clsd\n",
    "            \n",
    "    if vector_y_key is not None:\n",
    "        assert (primary_to_vector_y_transform is not None) == (vector_to_primary_y_transform is not None)\n",
    "        vector_y_axis_functions = (primary_to_vector_y_transform, vector_to_primary_y_transform) if (primary_to_vector_y_transform is not None) and (vector_to_primary_y_transform is not None) else None\n",
    "        vector_y_ax_clsd = ax_clsd.secondary_yaxis('right', functions=vector_y_axis_functions) # FIXME: primary x to vector x conversion and vice versa tuple for non-identity\n",
    "        if ax_open != ax_clsd:\n",
    "            vector_y_ax_open = ax_open.secondary_yaxis('right', functions=vector_y_axis_functions)\n",
    "        else:\n",
    "            vector_y_ax_open = vector_y_ax_clsd    \n",
    "\n",
    "    x_values_min = dict(closed=float('inf'), open=float('inf')) \n",
    "    x_values_max = dict(closed=float('-inf'), open=float('-inf'))\n",
    "    y_values_min = dict(closed=float('inf'), open=float('inf')) \n",
    "    y_values_max = dict(closed=float('-inf'), open=float('-inf'))\n",
    "\n",
    "    if vector_y_key is not None and vector_x_key is not None:\n",
    "        vector_y_values_range = [s[vector_y_key] if type(vector_y_key) is str else vector_y_key(s) for s in submissions_stats.values()]\n",
    "        vector_y_values_range = dict(min=np.min(vector_y_values_range), max=np.max(vector_y_values_range))\n",
    "        vector_x_values_range = [s[vector_x_key] if type(vector_x_key) is str else vector_x_key(s) for s in submissions_stats.values()]\n",
    "        vector_x_values_range = dict(min=np.min(vector_x_values_range), max=np.max(vector_x_values_range))\n",
    "        \n",
    "        vector_x_values_min = dict(closed=float('inf'), open=float('inf')) \n",
    "        vector_x_values_max = dict(closed=float('-inf'), open=float('-inf'))        \n",
    "        vector_y_values_min = dict(closed=float('inf'), open=float('inf')) \n",
    "        vector_y_values_max = dict(closed=float('-inf'), open=float('-inf'))\n",
    "        \n",
    "    # Plot all submissions\n",
    "    for i, (k, submission_stat) in enumerate(submissions_stats.items()):\n",
    "                \n",
    "        if not filter_submissions(submission_stat):\n",
    "            continue\n",
    "\n",
    "        submission_annotation = submission_stat[submission_annotation_key] if type(submission_annotation_key) is str else submission_annotation_key(submission_stat)\n",
    "        color = colors[k if color_label is None else color_label(submission_stat)]\n",
    "            \n",
    "        submission_stat_runs = sorted(submission_stat['runs'], key=lambda run: run['run_time'])\n",
    "        submission_stat_runs = submission_stat_runs[1:mlperf_opts[benchmark.round][benchmark.name]['runs_to_submit']-1] # olympic scoring rules\n",
    "\n",
    "        x_values = np.array([run[x_key] if type(x_key) is str else x_key({'run': run, 'system': submission_stat['system']}) for run in submission_stat_runs])\n",
    "        y_values = np.array([run[y_key] if type(y_key) is str else y_key({'run': run, 'system': submission_stat['system']}) for run in submission_stat_runs])\n",
    "\n",
    "        if x_transform is not None:\n",
    "            x_values = np.array([x_transform(val) for val in x_values])\n",
    "        x_values_min[submission_stat['division']] = min(x_values_min[submission_stat['division']], np.min(x_values))\n",
    "        x_values_max[submission_stat['division']] = max(x_values_max[submission_stat['division']], np.max(x_values))\n",
    "        if y_transform is not None:\n",
    "            y_values = np.array([y_transform(val) for val in y_values])\n",
    "        y_values_min[submission_stat['division']] = min(y_values_min[submission_stat['division']], np.min(y_values))\n",
    "        y_values_max[submission_stat['division']] = max(y_values_max[submission_stat['division']], np.max(y_values))\n",
    "\n",
    "        if vector_y_key is not None and vector_x_key is not None:\n",
    "            vector_x_values = submission_stat[vector_x_key] if type(vector_x_key) is str else vector_x_key(submission_stat)\n",
    "            vector_y_values = submission_stat[vector_y_key] if type(vector_y_key) is str else vector_y_key(submission_stat)\n",
    "\n",
    "            vector_x_values_min[submission_stat['division']] = min(vector_x_values_min[submission_stat['division']], vector_x_values/accelerators_to_epoch_throughput_ratio)\n",
    "            vector_x_values_max[submission_stat['division']] = max(vector_x_values_max[submission_stat['division']], vector_x_values/accelerators_to_epoch_throughput_ratio)\n",
    "            vector_y_values_min[submission_stat['division']] = min(vector_y_values_min[submission_stat['division']], vector_y_values/batch_size_to_epochs_ratio)\n",
    "            vector_y_values_max[submission_stat['division']] = max(vector_y_values_max[submission_stat['division']], vector_y_values/batch_size_to_epochs_ratio)\n",
    "        \n",
    "    \n",
    "        if submission_stat['division'] == 'closed':                \n",
    "            ax_clsd.scatter(x_values,\n",
    "                        y_values,\n",
    "                        s=5,\n",
    "                        color=color,\n",
    "                        label=submission_stat['submission_label'] + (f' ({color_label(submission_stat)})' if color_label is not None else ''),\n",
    "                        alpha=1.)\n",
    "            ax_clsd.annotate(submission_annotation,\n",
    "                        xy=(np.mean(x_values), np.mean(y_values)), \n",
    "                        color=color,\n",
    "                        **submission_label_annotation_opts)\n",
    "            if xy_mean_marker is not None:\n",
    "                ax_clsd.scatter(np.mean(x_values),\n",
    "                                np.mean(y_values),\n",
    "                                color=color,\n",
    "                                marker=xy_mean_marker,\n",
    "                                facecolor='white',\n",
    "                                alpha=1.)\n",
    "            if vector_y_key is not None and vector_x_key is not None:\n",
    "                ax.plot([np.mean(x_values), vector_to_primary_x_transform(vector_x_values)], [np.mean(y_values)]*2, get_directed_x_marker(np.mean(x_values), vector_to_primary_x_transform(vector_x_values)) + '-', markersize=12, markevery=[1], color=color) \n",
    "                ax.plot([np.mean(x_values)]*2, [np.mean(y_values), vector_to_primary_y_transform(vector_y_values)], get_directed_y_marker(np.mean(y_values), vector_to_primary_y_transform(vector_y_values)) + '-', markersize=12, markevery=[1], color=color) \n",
    "                if vector_annotate_label is not None:\n",
    "                    ax.annotate(vector_annotate_label(vector_x_values, vector_y_values, submission_stat), # r'$2^{' + str(int(np.log2(vector_y_values/vector_x_values))) + r'}$',\n",
    "                                xy=(vector_to_primary_x_transform(vector_x_values), vector_to_primary_y_transform(vector_y_values)),\n",
    "                                ha='center', va='center',  rotation=0,\n",
    "                                color=color) \n",
    "            # TODO: 2D-kde_estimate?\n",
    "        elif submission_stat['division'] == 'open':\n",
    "            ax_open.scatter(x_values,\n",
    "                        y_values,\n",
    "                        s=5,\n",
    "                        color=color,\n",
    "                        label=r'$\\bigstar$' + submission_stat['submission_label'] + (f' ({color_label(submission_stat)})' if color_label is not None else ''),\n",
    "                        alpha=1.)\n",
    "            ax_open.annotate(r'$\\bigstar$' + submission_annotation,\n",
    "                        xy=(np.mean(x_values), np.mean(y_values)),\n",
    "                        color=color,\n",
    "                        **submission_label_annotation_opts)        \n",
    "            if xy_mean_marker is not None:\n",
    "                ax_open.scatter(np.mean(x_values),\n",
    "                                np.mean(y_values),\n",
    "                                color=color,\n",
    "                                marker=xy_mean_marker,\n",
    "                                facecolor='white',\n",
    "                                alpha=1.)\n",
    "            if vector_y_key is not None and vector_x_key is not None:\n",
    "                ax.plot([np.mean(x_values), vector_to_primary_x_transform(vector_x_values)], [np.mean(y_values)]*2, get_directed_x_marker(np.mean(x_values), vector_to_primary_x_transform(vector_x_values)) + '-', markersize=12, markevery=[1], color=color) \n",
    "                ax.plot([np.mean(x_values)]*2, [np.mean(y_values), vector_to_primary_y_transform(vector_y_values)], get_directed_y_marker(np.mean(y_values), vector_to_primary_y_transform(vector_y_values)) + '-', markersize=12, markevery=[1], color=color) \n",
    "                if vector_annotate_label is not None:\n",
    "                    ax.annotate(vector_annotate_label(vector_x_values, vector_y_values, submission_stat), # r'$2^{' + str(int(np.log2(vector_y_values/vector_x_values))) + r'}$',\n",
    "                                xy=(vector_to_primary_x_transform(vector_x_values), vector_to_primary_y_transform(vector_y_values)),\n",
    "                                ha='center', va='center',  rotation=0,\n",
    "                                color=color) \n",
    "            # TODO: 2D-kde_estimate?\n",
    "        else:\n",
    "            raise ValueError(f\"Invalid division {submission_stat['division']}.\")\n",
    "\n",
    "    ax_clsd.legend(bbox_to_anchor=( 1.1 if ax_clsd == ax_open else -0.65, 1), loc='upper left', frameon=False, fontsize='small')\n",
    "    if ax_open != ax_clsd:\n",
    "        ax_open.legend(bbox_to_anchor=(1.55, 1), loc='upper right', frameon=False, fontsize='small')        \n",
    "\n",
    "    if vector_y_key is not None and vector_x_key is not None:    \n",
    "        vector_y_legend_marker = '^'\n",
    "        vector_x_legend_marker = '>'\n",
    "#         y_vector_legend_label = f'vertical vector for batch size, range = [{vector_y_values_range[\"min\"]},{vector_y_values_range[\"max\"]}]'\n",
    "        \n",
    "#         x_vector_legend_markers = ['s','<','>']\n",
    "#         x_vector_legend_labels = [f'= ref ({batch_size_to_accelerator_ratio})',\n",
    "#                                   f'< ref (min = {vector_angle_values_range[\"min\"]}) - more model-parallel',\n",
    "#                                   f'> ref (max = {vector_angle_values_range[\"max\"]}) - more data-parallel']\n",
    "        \n",
    "        symbolic_legend_labels = []\n",
    "        ax_clsd_twin = ax_clsd.twinx()\n",
    "        \n",
    "        # Symbolic legend\n",
    "        ax_clsd_twin.scatter([], [], label=\"Symbolic legend\", marker=\"None\") \n",
    "        symbolic_legend_labels.append(\"Symbolic legend\")\n",
    "        ax_clsd_twin.scatter([], [], label=x_legend_label, color='black')\n",
    "        symbolic_legend_labels.append(x_legend_label)\n",
    "        \n",
    "        ax_clsd_twin.scatter([], [], marker=vector_y_legend_marker, label=vector_y_legend_label, color='black')\n",
    "        symbolic_legend_labels.append(vector_y_legend_label)\n",
    "        ax_clsd_twin.scatter([], [], marker=vector_x_legend_marker, label=vector_x_legend_label, color='black')\n",
    "        symbolic_legend_labels.append(vector_x_legend_label)\n",
    "\n",
    "        # Color-coding\n",
    "        ax_clsd_twin.scatter([], [], label=\" \", marker=\"None\") \n",
    "        symbolic_legend_labels.append(\" \")\n",
    "        ax_clsd_twin.scatter([], [], label=\"Color-coding\", marker=\"None\") \n",
    "        symbolic_legend_labels.append(\"Color-coding\")\n",
    "        for color_key, color in colors.items():\n",
    "            ax_clsd_twin.scatter([], [], label=color_key, color=color, marker='o')\n",
    "            symbolic_legend_labels.append(color_key)\n",
    "\n",
    "        ax_clsd_twin.get_yaxis().set_visible(False)\n",
    "\n",
    "        legend_handles, legend_labels = ax_clsd_twin.get_legend_handles_labels()\n",
    "        legend_labels, legend_handles = zip(*sorted(zip(legend_labels, legend_handles), key=lambda t: symbolic_legend_labels.index(t[0])))\n",
    "        ax_clsd_twin.legend(legend_handles, legend_labels, bbox_to_anchor=( 1.6, 0), loc='lower right', frameon=False, fontsize='small')\n",
    "    \n",
    "    ax_clsd.grid(True, which='major', alpha=0.3)\n",
    "    ax_open.grid(True, which='major', alpha=0.3)\n",
    "\n",
    "    if axis_scale == 'linear':\n",
    "        ax_clsd.plot()\n",
    "        ax_open.plot()\n",
    "    elif axis_scale == 'semilogx':\n",
    "        ax_clsd.semilogx()\n",
    "        ax_open.semilogx()\n",
    "    elif axis_scale == 'loglog':\n",
    "        ax_clsd.loglog()\n",
    "        ax_open.loglog()\n",
    "    else:\n",
    "        raise ValueError(f'axis_scale should be in [semilogx, loglog, plot], not {axis_scale}')\n",
    "        \n",
    "    if vector_x_key is not None: # does the following have an effect at all?\n",
    "        vector_x_ax_clsd.xaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "        vector_x_ax_open.xaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "\n",
    "        vector_x_ax_clsd.xaxis.set_major_formatter(plt.FormatStrFormatter('%d'))\n",
    "        vector_x_ax_open.xaxis.set_major_formatter(plt.ScalarFormatter())\n",
    "        vector_x_ax_clsd.set_xlabel(vector_x_label if vector_x_label is not None else vector_x_key)\n",
    "        vector_x_ax_open.set_xlabel(vector_x_label if vector_x_label is not None else vector_x_key)    \n",
    "    \n",
    "    if vector_y_key is not None: # does the following have an effect at all?\n",
    "        vector_y_ax_clsd.yaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "        vector_y_ax_open.yaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "\n",
    "        vector_y_ax_clsd.yaxis.set_major_formatter(plt.FormatStrFormatter('%d'))\n",
    "        vector_y_ax_open.yaxis.set_major_formatter(plt.ScalarFormatter())\n",
    "        vector_y_ax_clsd.set_ylabel(vector_y_label if vector_y_label is not None else vector_y_key)\n",
    "        vector_y_ax_open.set_ylabel(vector_y_label if vector_y_label is not None else vector_y_key)    \n",
    "    \n",
    "    if level_func is not None:\n",
    "        if level_transform is not None:\n",
    "            if len(inspect.signature(level_transform[0]).parameters) > 1: # submission-dependent transformation\n",
    "                level_transform_partial = (functools.partial(level_transform[0], s=submission_stat),\n",
    "                                           functools.partial(level_transform[1], s=submission_stat))\n",
    "            else:\n",
    "                level_transform_partial = level_transform\n",
    "\n",
    "        plot_level_lines(ax=ax_clsd, \n",
    "                        x_range=(min(x_values_min['closed'],x_values_min['open']), max(x_values_max['closed'],x_values_max['open'])), \n",
    "                        y_range=(min(y_values_min['closed'],y_values_min['open']), max(y_values_max['closed'],y_values_max['open'])), \n",
    "                        level_func=level_func, \n",
    "                        level_label=level_label,\n",
    "                        level_transform=level_transform_partial if level_transform is not None else None,\n",
    "                        level_log_base=level_log_base,\n",
    "                        level_fine_grid=level_fine_grid,\n",
    "                        level_color=level_color)\n",
    "\n",
    "    if vector_level_func is not None:\n",
    "        if vector_level_transform is not None:\n",
    "            if len(inspect.signature(vector_level_transform[0]).parameters) > 1: # submission-dependent transformation\n",
    "                vector_level_transform_partial = (functools.partial(vector_level_transform[0], s=submission_stat),\n",
    "                                           functools.partial(vector_level_transform[1], s=submission_stat))\n",
    "            else:\n",
    "                vector_level_transform_partial = vector_level_transform\n",
    "\n",
    "        plot_level_lines(ax=ax_clsd, \n",
    "                        x_range=(min(vector_x_values_min['closed'],vector_x_values_min['open']), max(vector_x_values_max['closed'],vector_x_values_max['open'])), \n",
    "                        y_range=(min(vector_y_values_min['closed'],vector_y_values_min['open']), max(vector_y_values_max['closed'],vector_y_values_max['open'])), \n",
    "                        level_func=vector_level_func, \n",
    "                        level_label=vector_level_label,\n",
    "                        level_transform=vector_level_transform_partial if vector_level_transform is not None else None,\n",
    "                        level_log_base=vector_level_log_base,\n",
    "                        level_color=vector_level_color)\n",
    "    \n",
    "    # No axis limits set\n",
    "\n",
    "    ax_clsd.set_xlabel(x_label if x_label is not None else x_key)\n",
    "    ax_open.set_xlabel(x_label if x_label is not None else x_key)\n",
    "    ax_clsd.set_ylabel(y_label)\n",
    "    ax_open.set_ylabel(y_label)\n",
    "\n",
    "    if fig is not None:\n",
    "#         fig.suptitle(benchmark.round + '/' + benchmark.name + ' - '+ (title if title is not None else f'{x_label} vs. {y_label}'))\n",
    "        fig.tight_layout()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Label definitions/helpers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "if benchmark.round == 'hpc_results_v0.7':\n",
    "    accelerator_short_names = [\n",
    "        'A64FX',\n",
    "        'Intel Xeon Phi',\n",
    "        'P100',\n",
    "        'Quadro RTX 5000',\n",
    "        'V100']\n",
    "    color_legend_label = {\n",
    "        'A64FX': 'Fujitsu A64FX',\n",
    "        'V100': 'NVIDIA V100',\n",
    "        'P100': 'NVIDIA P100',\n",
    "        'Intel Xeon Phi': 'Intel Xeon Phi 7250',\n",
    "        'Quadro RTX 5000': 'NVIDIA Quadro RTX 5000',\n",
    "    }\n",
    "\n",
    "\n",
    "def accelerator_name(system):\n",
    "    return system['accelerator_model_name'] if int(system['accelerators_per_node']) > 0 else system['host_processor_model_name'] \n",
    "\n",
    "\n",
    "def accelerator_short_label(submission_stat):\n",
    "    name = accelerator_name(submission_stat['system']) if not isinstance(submission_stat, str) else submission_stat\n",
    "    for short_name in accelerator_short_names:\n",
    "        if short_name in name:\n",
    "            return short_name\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "show_submission_colors(color_label=accelerator_short_label, color_label_range=accelerator_short_names, alpha=0.3)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Plot parameters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Plot parameters\n",
    "\n",
    "# x-axis\n",
    "\n",
    "x_label = 'batch size'\n",
    "x_key = 'global_batch_size'\n",
    "\n",
    "x_label = '# accelerators'\n",
    "x_key = lambda submission_stat: accelerator_number(submission_stat['system'])\n",
    "\n",
    "# y-axis\n",
    "\n",
    "y_label = '# epochs to solution'\n",
    "y_key = 'run_stop_epoch'\n",
    "y_transform = None\n",
    "\n",
    "y_label = 'time to solution [min]'\n",
    "y_key = 'run_time'\n",
    "y_transform = lambda t: t/60.\n",
    "    \n",
    "y_label = 'epoch throughput [1/min]'\n",
    "y_key = 'run_epoch_throughput'\n",
    "y_transform = lambda t: t/60.\n",
    "    \n",
    "def get_accelerator_hours(run_system_stats):\n",
    "    return run_system_stats['run']['run_time']/3600*accelerator_number(run_system_stats['system'])\n",
    "\n",
    "y_label = 'accelerator hours'\n",
    "y_key = get_accelerator_hours\n",
    "y_transform = None"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1.0"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "if benchmark.round == 'hpc_results_v0.7':\n",
    "    batch_size_to_accelerator_ratio = 1.\n",
    "batch_size_to_accelerator_ratio"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Individual hyperparameter effects (off)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Checking for correlations of individual hyperparameters on number of epochs to converge is possible \n",
    "# (not very helpful when everyone roughly uses the same - hence \"if False\")\n",
    "\n",
    "if benchmark.name == 'cosmoflow' and False:\n",
    "    x_key = lambda run_system_stats: float(run_system_stats['opt_opts']['opt_learning_rate_decay_boundary_epochs'][1])\n",
    "    x_label = 'opt_learning_rate_decay_boundary_epochs[1]'\n",
    "    filter_submissions = lambda submission_stat: submission_stat['division'] == 'closed'\n",
    "\n",
    "    y_label = '# epochs to solution'\n",
    "    y_key = 'run_stop_epoch'\n",
    "    y_transform = None\n",
    "\n",
    "    plot_all_submissions(x_key=x_key, x_label=x_label, y_key=y_key, y_label=y_label, y_transform=y_transform, sharey=True, filter_submissions=filter_submissions)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Time to solution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "def generate_time_to_solution_plot():\n",
    "    fig, ax = plt.subplots(ncols=1,\n",
    "                           figsize=(18, 12))\n",
    "\n",
    "    # x_label = 'batch size'\n",
    "    # x_key = 'global_batch_size'\n",
    "\n",
    "    x_label = '# compute units'\n",
    "    x_key = lambda submission_stat: accelerator_number(submission_stat['system'])\n",
    "\n",
    "    secondary_x_label = 'batch size'\n",
    "    secondary_x_legend_label = 'time to solution [min] (mean) at batch size (-- to # compute units)'\n",
    "    secondary_x_key = 'global_batch_size'\n",
    "    secondary_x_marker = 'd'\n",
    "\n",
    "    primary_to_secondary_x_transform = lambda a: a*batch_size_to_accelerator_ratio\n",
    "    secondary_to_primary_x_transform = lambda b: b/batch_size_to_accelerator_ratio\n",
    "\n",
    "    y_label = 'time to solution [min]'\n",
    "    y_legend_label = 'time to solution [min] (dist) at # compute units'\n",
    "    y_key = 'run_time'\n",
    "    y_transform = lambda t: t/60.\n",
    "\n",
    "    plot_all_submissions(x_key=x_key, x_label=x_label, \n",
    "                         secondary_x_key=secondary_x_key, secondary_x_label=secondary_x_label, secondary_x_legend_label=secondary_x_legend_label, secondary_x_marker=secondary_x_marker,\n",
    "                         y_key=y_key, y_label=y_label, y_legend_label=y_legend_label, y_transform=y_transform, \n",
    "                         primary_to_secondary_x_transform=primary_to_secondary_x_transform, secondary_to_primary_x_transform=secondary_to_primary_x_transform,\n",
    "                         sharey=True, axis_scale='loglog', color_label=accelerator_short_label, color_label_range=accelerator_short_names, show_symbolic_legend=True,\n",
    "                         level_func='x*y', level_label=(r'compute budget', r'$[compute unit \\cdot h]$'), level_transform=(lambda l, s: l/60, lambda l, s: l*60), ax=ax)\n",
    "\n",
    "    ax.yaxis.set_major_locator(plt.LogLocator(base=10.))\n",
    "    ax.yaxis.set_major_formatter(plt.FormatStrFormatter('%d'))\n",
    "    \n",
    "# generate_time_to_solution_plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_effective_epoch_samples(submissions_stats):\n",
    "    submission_train_samples = np.unique([stat['train_samples'] for stat in submissions_stats.values()])\n",
    "    if len(submission_train_samples) > 1:\n",
    "        print(\"Note: different number of training samples used in different submissions! Using median of {}\".format(submission_train_samples))\n",
    "    submission_train_samples = np.median(submission_train_samples)\n",
    "\n",
    "\n",
    "    submission_eval_samples = np.unique([stat['eval_samples'] for stat in submissions_stats.values()])\n",
    "    if len(submission_eval_samples) > 1:\n",
    "        print(\"Note: different number of evaluation samples used in different submissions! Using median of {}\".format(submission_train_samples))\n",
    "    submission_eval_samples = np.median(submission_eval_samples)\n",
    "\n",
    "    mean_eff_epoch_samples = [] # handle DeepCAM separately\n",
    "    submissions_runs_with_olympic_scoring = [(k, sorted(submission_stat['runs'], key=lambda run: run['run_time'])[1:mlperf_opts[benchmark.round][benchmark.name]['runs_to_submit']-1]) \n",
    "                                             for k, submission_stat in submissions_stats.items()] # olympic scoring rules\n",
    "    for k, runs in submissions_runs_with_olympic_scoring:\n",
    "        submission_eff_epoch_samples = [runs[i]['run_avg_epoch_samples'] for i in range(len(runs))]\n",
    "        if benchmark.round == 'hpc_results_v0.7' and benchmark.name == 'cosmoflow':\n",
    "            assert np.unique(submission_eff_epoch_samples) == np.array([submission_train_samples + submission_eval_samples])\n",
    "        else:\n",
    "            mean_eff_epoch_samples.append(np.mean(submission_eff_epoch_samples))\n",
    "    \n",
    "    if benchmark.round == 'hpc_results_v0.7' and benchmark.name == 'cosmoflow':\n",
    "        eff_epoch_samples = submission_train_samples + submission_eval_samples\n",
    "    else:\n",
    "        eff_epoch_samples = int(np.mean(mean_eff_epoch_samples))\n",
    "    return eff_epoch_samples\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1000.0"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "if benchmark.round == 'hpc_results_v0.7':\n",
    "    accelerators_to_epoch_throughput_ratio = 1000./1.\n",
    "accelerators_to_epoch_throughput_ratio"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "5.12"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "if benchmark.round == 'hpc_results_v0.7':\n",
    "    batch_size_to_epochs_ratio = 256/50.\n",
    "batch_size_to_epochs_ratio\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Compute time analysis: Parameter choice (batch size, # compute units), epoch- and throughput scaling, compute time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "def generate_compute_time_analysis_plot():\n",
    "    gridspec_kw = dict(width_ratios=(1,1), height_ratios=(1,10/7))\n",
    "    fig, ax = plt.subplots(nrows=2, ncols=2, \n",
    "                           figsize=(32, 17), gridspec_kw=gridspec_kw)\n",
    "\n",
    "    if benchmark.round == 'hpc_results_v0.7':\n",
    "        filter_submissions = lambda s: True # e.g. s['system']['submitter'] in ['CSCS', 'Fujitsu', 'Fujitsu/RIKEN', 'LBNL', 'NCSA', 'TACC']\n",
    "\n",
    "    #### 0,0\n",
    "\n",
    "    y_label = 'batch size'\n",
    "    y_key = 'global_batch_size'\n",
    "\n",
    "    x_label = '# compute units'\n",
    "    x_key = lambda submission_stat: accelerator_number(submission_stat['system'])\n",
    "\n",
    "    def submission_label_annotation_opts_parameter_choice(submission_stat):\n",
    "        label = submission_stat['submission_long_label']\n",
    "        if benchmark.name == 'cosmoflow':\n",
    "            if label.endswith('-512'):\n",
    "                y_coord = {'Cori-KNL-512': -30, 'Fugaku-512': 0, 'ABCI-512': 30}\n",
    "                return dict(xytext=( -8, y_coord[label]), textcoords=\"offset points\", ha='right', va='center', rotation=0, alpha=1., fontsize=12)        \n",
    "            elif label.endswith('-64'):\n",
    "                y_coord = {'HAL-64': -30, 'Cori-GPU-64': 0, 'Frontera\\n-RTX-64': 30}\n",
    "                return dict(xytext=( -8, y_coord[label]), textcoords=\"offset points\", ha='right', va='center', rotation=0, alpha=1., fontsize=12)  \n",
    "            else:\n",
    "                return dict(xytext=( -8, 0), textcoords=\"offset points\", ha='right', va='center', rotation=0, alpha=1., fontsize=12)  \n",
    "        elif benchmark.name == 'deepcam':\n",
    "            if label.endswith('-1024'):\n",
    "                return dict(xytext=( -8,15 if submission_stat['division'] == 'open' else -15), textcoords=\"offset points\", ha='right', va='center', rotation=0, alpha=1., fontsize=12)        \n",
    "            else:\n",
    "                return dict(xytext=( -8, 0), textcoords=\"offset points\", ha='right', va='center', rotation=0, alpha=1., fontsize=12)  \n",
    "        else:\n",
    "            return dict(xytext=( -8, 0), textcoords=\"offset points\", ha='right', va='center', rotation=0, alpha=1., fontsize=12)  \n",
    "\n",
    "    submission_label_annotation_opts = submission_label_annotation_opts_parameter_choice\n",
    "\n",
    "    plot_all_submissions_flipped(x_key=x_key, x_label=x_label,\n",
    "                         y_key=y_key, y_label=y_label, \n",
    "                         level_func='y/x', level_label=(r'$\\dfrac{batch\\; size}{\\#\\; compute\\; units}$', '(communication intensity:\\n' + r'$\\nwarrow$ low, $\\searrow$ high)'), level_log_base=10., \n",
    "                         sharey=True, axis_scale='loglog', color_label=accelerator_short_label, color_label_range=accelerator_short_names,\n",
    "                         format_x_axis_as_y_axis=True, \n",
    "                         filter_submissions=filter_submissions,\n",
    "                         submission_annotation_key='submission_long_label' if benchmark.name == 'deepcam' else 'submission_short_label', \n",
    "                         submission_label_annotation_opts=submission_label_annotation_opts, ax=ax[0,0])\n",
    "\n",
    "    #### 1,0\n",
    "\n",
    "    secondary_x_label = r'batch size ($\\diamond$)' + '\\n '\n",
    "    secondary_x_legend_label = 'epoch throughput [1/min] (mean) at batch size (-- to # acc)'\n",
    "    secondary_x_key = 'global_batch_size'\n",
    "    secondary_x_marker = 'D' \n",
    "\n",
    "    x_label = r'# compute units' #r'# compute units ($\\blacktriangle, \\times, \\circ$' + (')' if benchmark.name == 'cosmoflow' else r', $\\mathbf{+}$)')\n",
    "    x_key = lambda submission_stat: accelerator_number(submission_stat['system'])\n",
    "\n",
    "    primary_to_secondary_x_transform = lambda a: a*batch_size_to_accelerator_ratio\n",
    "    secondary_to_primary_x_transform = lambda b: b/batch_size_to_accelerator_ratio\n",
    "\n",
    "    # TODO: epoch time\n",
    "\n",
    "    y_label = r'epoch throughput [1/min] ($\\circ$)'\n",
    "    y_legend_label = 'epoch throughput [1/min] (dist) at # compute units = combined throughput [samples/s]' if benchmark.round == 'hpc_results_v0.7' and benchmark.name == 'cosmoflow' else 'epoch throughput [1/min] (dist) at # acc'\n",
    "    y_key = 'run_epoch_throughput'\n",
    "    y_transform = lambda t: t*60.\n",
    "\n",
    "    secondary_y_keys = []\n",
    "    secondary_y_legend_labels = []\n",
    "    secondary_y_transforms = []\n",
    "    secondary_y_markers = []\n",
    "\n",
    "    secondary_y_keys.append('run_train_throughput')\n",
    "    secondary_y_transforms.append(lambda t: t) #*60.\n",
    "    secondary_y_markers.append('^')\n",
    "    secondary_y_legend_labels.append(r'training throughput [samples/s] (mean$\\pm$std) at # acc')\n",
    "\n",
    "    secondary_y_keys.append('run_eval_throughput')\n",
    "    secondary_y_transforms.append(lambda t: t) #*60.\n",
    "    secondary_y_markers.append('x')\n",
    "    secondary_y_legend_labels.append(r'evaluation throughput [samples/s] (mean$\\pm$std) at # acc')\n",
    "\n",
    "    if not(benchmark.round == 'hpc_results_v0.7' and benchmark.name == 'cosmoflow'):\n",
    "        secondary_y_keys.append('run_avg_epoch_sample_throughput')\n",
    "        secondary_y_transforms.append(lambda t: t) #*60.\n",
    "        secondary_y_markers.append('P')\n",
    "        secondary_y_legend_labels.append(r'average throughput [samples/s] (mean$\\pm$std) at # acc')\n",
    "\n",
    "    # if benchmark.round == 'hpc_results_v0.7':\n",
    "    #     secondary_y_keys.append('run_staging_throughput')\n",
    "    #     secondary_y_transforms.append(lambda t: t) #*60.\n",
    "    #     secondary_y_markers.append('s')\n",
    "    #     secondary_y_legend_labels.append(r'staging throughput [samples/s] (mean$\\pm$std) at # acc')\n",
    "\n",
    "    secondary_y_axis_label = 'sample throughput [samples/s]\\n' + r'($\\blacktriangle$ training, $\\times$ evaluation, ' + (r'$\\circ$' if benchmark.name == 'cosmoflow' else r'$\\mathbf{+} $') + ' combined)'\n",
    "\n",
    "    eff_epoch_samples = get_effective_epoch_samples(submissions_stats)\n",
    "    primary_to_secondary_y_transform = lambda et: et*float(eff_epoch_samples)/60. \n",
    "    secondary_to_primary_y_transform = lambda st: st/float(eff_epoch_samples)*60.\n",
    "\n",
    "    # Epoch throughput w/o initial phase, i.e. staging overhead not shown\n",
    "\n",
    "    level_func='y/x' \n",
    "    level_label=('per compute unit\\nsample throughput\\n' + r'$[samples/s]$', r'($\\nwarrow$ high, $\\searrow$ low)')\n",
    "    level_transform=(primary_to_secondary_y_transform, secondary_to_primary_y_transform)\n",
    "\n",
    "    plot_all_submissions(y_key=y_key, y_label=y_label, y_transform=y_transform, y_legend_label=y_legend_label, y_mean_marker='o',\n",
    "                         secondary_y_keys=secondary_y_keys, secondary_y_axis_label=secondary_y_axis_label, secondary_y_legend_labels=secondary_y_legend_labels, secondary_y_transforms=secondary_y_transforms, secondary_y_markers=secondary_y_markers,\n",
    "                         primary_to_secondary_y_transform=primary_to_secondary_y_transform, secondary_to_primary_y_transform=secondary_to_primary_y_transform,\n",
    "                         x_key=x_key, x_label=x_label,\n",
    "                         #secondary_x_key=secondary_x_key, secondary_x_label=secondary_x_label, secondary_x_legend_label=secondary_x_legend_label, secondary_x_marker=secondary_x_marker, secondary_x_ticks_same_as_primary=True,\n",
    "                         #primary_to_secondary_x_transform=primary_to_secondary_x_transform, secondary_to_primary_x_transform=secondary_to_primary_x_transform, \n",
    "                         sharey=False, axis_scale='loglog', color_label=accelerator_short_label, color_label_range=accelerator_short_names, legend_side='left', show_symbolic_legend=True,\n",
    "                         level_func=level_func, level_label=level_label, level_transform=level_transform, level_fine_grid=2, filter_submissions=filter_submissions, \n",
    "                         submission_annotation_key='submission_long_label' if benchmark.name == 'deepcam' else 'submission_short_label', \n",
    "                         color_legend_label=color_legend_label, ax=ax[1,0])        \n",
    "\n",
    "    ### 0,1\n",
    "\n",
    "    y_label = 'batch size'\n",
    "    y_key = 'global_batch_size'\n",
    "\n",
    "    x_label = '# epochs to solution'\n",
    "    x_key = 'run_stop_epoch'\n",
    "    x_transform = None\n",
    "\n",
    "    def submission_label_annotation_opts_epoch_scaling(submission_stat):\n",
    "        label = submission_stat['submission_long_label']\n",
    "        if benchmark.name == 'cosmoflow':\n",
    "            if label == 'Piz-Daint-256':\n",
    "                return dict(xytext=( 10, -2),\n",
    "                            textcoords=\"offset points\",\n",
    "                            ha='center', va='top', rotation=90,      \n",
    "                            alpha=1., fontsize=12)        \n",
    "            elif label in ['HAL-64']:\n",
    "                return dict(xytext=( 13, -2),\n",
    "                            textcoords=\"offset points\",\n",
    "                            ha='center', va='top', rotation=90,      \n",
    "                            alpha=1., fontsize=12)        \n",
    "            elif label in ['Fugaku-512']:\n",
    "                return dict(xytext=( -15, -2),\n",
    "                            textcoords=\"offset points\",\n",
    "                            ha='center', va='top', rotation=90,      \n",
    "                            alpha=1., fontsize=12)        \n",
    "            elif label in ['Fugaku-8192']:\n",
    "                return dict(xytext=( 5, -2),\n",
    "                            textcoords=\"offset points\",\n",
    "                            ha='center', va='top', rotation=90,      \n",
    "                            alpha=1., fontsize=12)        \n",
    "            elif 'Frontera' in label:\n",
    "                return dict(xytext=( -5, -2),\n",
    "                            textcoords=\"offset points\",\n",
    "                            ha='center', va='top', rotation=90,      \n",
    "                            alpha=1., fontsize=12)        \n",
    "            else:\n",
    "                return dict(xytext=( 0, -2),\n",
    "                            textcoords=\"offset points\",\n",
    "                            ha='center', va='top', rotation=90,      \n",
    "                            alpha=1., fontsize=12)\n",
    "        elif benchmark.name == 'deepcam':\n",
    "            if label.endswith('-1024'):\n",
    "                return dict(xytext=( -3 if submission_stat['division'] == 'open' else 3, -2.5),\n",
    "                            textcoords=\"offset points\",\n",
    "                            ha='center', va='top', rotation=90,      \n",
    "                            alpha=1., fontsize=12)        \n",
    "            else:\n",
    "                return dict(xytext=( 0, -2.5),\n",
    "                            textcoords=\"offset points\",\n",
    "                            ha='center', va='top', rotation=90,      \n",
    "                            alpha=1., fontsize=12)\n",
    "        else:\n",
    "            return dict(xytext=( 0, -2.5),\n",
    "                        textcoords=\"offset points\",\n",
    "                        ha='center', va='top', rotation=90,      \n",
    "                        alpha=1., fontsize=12)\n",
    "\n",
    "    submission_label_annotation_opts = submission_label_annotation_opts_epoch_scaling\n",
    "    plot_all_submissions_flipped(y_key=y_key, y_label=y_label, \n",
    "                         x_key=x_key, x_label=x_label, x_transform=x_transform, \n",
    "                         level_func='x/y', level_label=('# train steps to solution', r'($\\nwarrow$ low, $\\searrow$ high)'), \n",
    "                         level_transform=(lambda l, s: l*s['train_samples'], lambda l, s: l/s['train_samples']), level_fine_grid=3,\n",
    "                         sharey=True, axis_scale='loglog', color_label=accelerator_short_label, color_label_range=accelerator_short_names, filter_submissions=filter_submissions, \n",
    "                         submission_annotation_key='submission_epochs_label',\n",
    "                         submission_label_annotation_opts=submission_label_annotation_opts, ax=ax[0,1])\n",
    "\n",
    "    ax[0,1].xaxis.set_major_locator(plt.AutoLocator())\n",
    "    ax[0,1].xaxis.set_major_formatter(plt.FormatStrFormatter('%d'))\n",
    "    ax[0,1].yaxis.set_major_locator(plt.LogLocator(base=2.))\n",
    "    ax[0,1].yaxis.set_major_formatter(plt.FormatStrFormatter('%d'))\n",
    "\n",
    "    ### 1,1\n",
    "\n",
    "    x_label = '# epochs to solution'\n",
    "    x_key = 'run_stop_epoch'\n",
    "    x_transform = None\n",
    "\n",
    "    y_label = 'epoch throughput [1/min]'\n",
    "    y_key = 'run_epoch_throughput'\n",
    "    y_transform = lambda t: t*60.\n",
    "\n",
    "    level_func='x/y' \n",
    "    level_label=('compute time\\nto solution' + r' $[min]$', r'($\\nwarrow$ low, $\\searrow$ high)') \n",
    "\n",
    "    submission_label_annotation_opts = dict(\n",
    "                        xytext=(0, 0),\n",
    "                        textcoords=\"offset points\",\n",
    "                        ha='right', va='top',  rotation=14.75,\n",
    "                        alpha=1., fontsize=12)\n",
    "\n",
    "    plot_all_submissions_scatter_vector(x_key=x_key, x_label=x_label, x_transform=x_transform, \n",
    "                                        y_key=y_key, y_label=y_label, y_transform=y_transform,\n",
    "                                        xy_mean_marker='o',\n",
    "                                        sharey=True, axis_scale='loglog', color_label=accelerator_short_label, color_label_range=accelerator_short_names,\n",
    "                                        level_func=level_func, level_label=level_label, level_fine_grid=2,\n",
    "                                        filter_submissions=filter_submissions,\n",
    "                                        submission_annotation_key='submission_long_label',\n",
    "                                        submission_label_annotation_opts=submission_label_annotation_opts,\n",
    "                                        ax=ax[1,1])\n",
    "\n",
    "    ax[1,1].xaxis.set_major_locator(plt.AutoLocator())\n",
    "    ax[1,1].xaxis.set_major_formatter(plt.FormatStrFormatter('%d'))\n",
    "\n",
    "    ax[0,0].set_xlim(ax[1,0].get_xlim())\n",
    "    ax[0,0].set_ylim(ax[0,1].get_ylim())\n",
    "\n",
    "    ax[1,1].set_xlim(ax[0,1].get_xlim())\n",
    "    ax[1,1].set_ylim(ax[1,0].get_ylim())\n",
    "\n",
    "    ax[0,0].legend().set_visible(False)\n",
    "    ax[0,1].legend().set_visible(False)\n",
    "    ax[1,0].legend().set_visible(False)\n",
    "    ax[1,1].legend().set_visible(False)\n",
    "\n",
    "    fig.tight_layout()\n",
    "# generate_compute_time_analysis_plot()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Tabular analysis"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "def olympic_scoring_of_runs(runs):\n",
    "    return sorted(runs, key=lambda run: run['run_time'])[1:mlperf_opts[benchmark.round][benchmark.name]['runs_to_submit']-1] # olympic scoring rules (discards the best and worst run (if invalid should be skipped at parsing))\n",
    "\n",
    "from copy import deepcopy\n",
    "submissions_stats_olympic = deepcopy(submissions_stats)\n",
    "for k in submissions_stats_olympic:\n",
    "    submissions_stats_olympic[k]['runs'] = olympic_scoring_of_runs(submissions_stats_olympic[k]['runs']) # olympic scoring rules"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "submissions_stats_df = pd.DataFrame.from_dict(submissions_stats_olympic, orient='index').sort_values(['division'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "pd.set_option('display.max_columns', None)\n",
    "def visualize_submissions_by_opt(submissions_stats_df_opt_vis, sort_keys=['global_batch_size']):\n",
    "    if benchmark.round == 'training_results_v0.7':\n",
    "        display(submissions_stats_df_opt_vis[(submissions_stats_df_opt_vis['division'] == 'closed') & (submissions_stats_df_opt_vis['opt_name'] == 'sgd')].drop(columns=[c for c in submissions_stats_df_opt_vis.columns if c.startswith('lars_')]).sort_values(sort_keys))\n",
    "        display(submissions_stats_df_opt_vis[(submissions_stats_df_opt_vis['division'] == 'closed') & (submissions_stats_df_opt_vis['opt_name'] == 'lars')].drop(columns=[c for c in submissions_stats_df_opt_vis.columns if c.startswith('sgd_') or c == 'model_bn_span' or c == 'opt_learning_rate_warmup_epochs']).sort_values(sort_keys))\n",
    "        display(submissions_stats_df_opt_vis[(submissions_stats_df_opt_vis['division'] == 'open')].sort_values(sort_keys))\n",
    "    else:\n",
    "        display(submissions_stats_df_opt_vis.sort_values(['division'] + sort_keys))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Runtime statistics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>global_batch_size</th>\n",
       "      <th>opt_name</th>\n",
       "      <th>train_samples</th>\n",
       "      <th>eval_samples</th>\n",
       "      <th>division</th>\n",
       "      <th>submission_long_label</th>\n",
       "      <th>submission_epochs_label</th>\n",
       "      <th>eval_time/run_time</th>\n",
       "      <th>run_avg_epoch_sample_throughput.mean</th>\n",
       "      <th>run_avg_epoch_sample_throughput.std</th>\n",
       "      <th>run_avg_epoch_samples.mean</th>\n",
       "      <th>run_avg_epoch_samples.std</th>\n",
       "      <th>run_epoch_throughput.mean</th>\n",
       "      <th>run_epoch_throughput.std</th>\n",
       "      <th>run_epoch_time.mean</th>\n",
       "      <th>run_epoch_time.std</th>\n",
       "      <th>run_eval_throughput.mean</th>\n",
       "      <th>run_eval_throughput.std</th>\n",
       "      <th>run_eval_time.mean</th>\n",
       "      <th>run_eval_time.std</th>\n",
       "      <th>run_num_evals.mean</th>\n",
       "      <th>run_num_evals.std</th>\n",
       "      <th>run_staging_throughput.mean</th>\n",
       "      <th>run_staging_throughput.std</th>\n",
       "      <th>run_staging_time.mean</th>\n",
       "      <th>run_staging_time.std</th>\n",
       "      <th>run_stop_epoch.mean</th>\n",
       "      <th>run_stop_epoch.std</th>\n",
       "      <th>run_time.mean</th>\n",
       "      <th>run_time.std</th>\n",
       "      <th>run_train_throughput.mean</th>\n",
       "      <th>run_train_throughput.std</th>\n",
       "      <th>run_train_time.mean</th>\n",
       "      <th>run_train_time.std</th>\n",
       "      <th>staging_time/run_time</th>\n",
       "      <th>train_time/run_time</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>submission_label</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>LBNL-1/corigpu_n64_pt1.6.0</th>\n",
       "      <td>128</td>\n",
       "      <td>LAMB</td>\n",
       "      <td>121216</td>\n",
       "      <td>15158</td>\n",
       "      <td>closed</td>\n",
       "      <td>Cori-GPU-64</td>\n",
       "      <td>Cori\\n-GPU-64</td>\n",
       "      <td>0.362732</td>\n",
       "      <td>13636.646533</td>\n",
       "      <td>307.016017</td>\n",
       "      <td>189932.0</td>\n",
       "      <td>874.685658</td>\n",
       "      <td>0.071793</td>\n",
       "      <td>0.001919</td>\n",
       "      <td>13.928849</td>\n",
       "      <td>0.362650</td>\n",
       "      <td>13605.274039</td>\n",
       "      <td>704.124769</td>\n",
       "      <td>50.524406</td>\n",
       "      <td>3.044168</td>\n",
       "      <td>45.333333</td>\n",
       "      <td>0.577350</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>10.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>139.288611</td>\n",
       "      <td>3.626479</td>\n",
       "      <td>13655.973840</td>\n",
       "      <td>496.046978</td>\n",
       "      <td>88.764083</td>\n",
       "      <td>3.111298</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.637267</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fujitsu-1/abci_1024xV100_pytorch_closed</th>\n",
       "      <td>2048</td>\n",
       "      <td>LAMB</td>\n",
       "      <td>121266</td>\n",
       "      <td>15158</td>\n",
       "      <td>closed</td>\n",
       "      <td>ABCI-1024</td>\n",
       "      <td>ABCI-1024</td>\n",
       "      <td>0.040034</td>\n",
       "      <td>328452.837177</td>\n",
       "      <td>954.164474</td>\n",
       "      <td>130108.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>2.524463</td>\n",
       "      <td>0.007334</td>\n",
       "      <td>0.396124</td>\n",
       "      <td>0.001147</td>\n",
       "      <td>452842.370981</td>\n",
       "      <td>427.222282</td>\n",
       "      <td>0.468622</td>\n",
       "      <td>0.000442</td>\n",
       "      <td>14.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>62072.354483</td>\n",
       "      <td>180.453773</td>\n",
       "      <td>2.197822</td>\n",
       "      <td>0.006371</td>\n",
       "      <td>24.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>11.705717</td>\n",
       "      <td>0.021555</td>\n",
       "      <td>322003.905651</td>\n",
       "      <td>993.958276</td>\n",
       "      <td>9.038350</td>\n",
       "      <td>0.027814</td>\n",
       "      <td>0.187756</td>\n",
       "      <td>0.772131</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fujitsu-2/abci_1024xV100_fjpytorch_open</th>\n",
       "      <td>2048</td>\n",
       "      <td>LAMB</td>\n",
       "      <td>121266</td>\n",
       "      <td>15158</td>\n",
       "      <td>open</td>\n",
       "      <td>ABCI-1024</td>\n",
       "      <td>ABCI\\n-1024</td>\n",
       "      <td>0.013057</td>\n",
       "      <td>341240.183868</td>\n",
       "      <td>8540.870875</td>\n",
       "      <td>122953.0</td>\n",
       "      <td>639.126748</td>\n",
       "      <td>2.775585</td>\n",
       "      <td>0.056186</td>\n",
       "      <td>0.360284</td>\n",
       "      <td>0.007149</td>\n",
       "      <td>281674.493324</td>\n",
       "      <td>53363.772466</td>\n",
       "      <td>0.136906</td>\n",
       "      <td>0.033568</td>\n",
       "      <td>2.666667</td>\n",
       "      <td>1.154701</td>\n",
       "      <td>69521.318153</td>\n",
       "      <td>2846.825395</td>\n",
       "      <td>1.962333</td>\n",
       "      <td>0.077194</td>\n",
       "      <td>23.666667</td>\n",
       "      <td>1.154701</td>\n",
       "      <td>10.485394</td>\n",
       "      <td>0.228172</td>\n",
       "      <td>342041.745864</td>\n",
       "      <td>8021.315726</td>\n",
       "      <td>8.385078</td>\n",
       "      <td>0.246814</td>\n",
       "      <td>0.187149</td>\n",
       "      <td>0.799691</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                         global_batch_size opt_name  \\\n",
       "submission_label                                                      \n",
       "LBNL-1/corigpu_n64_pt1.6.0                             128     LAMB   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed               2048     LAMB   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open               2048     LAMB   \n",
       "\n",
       "                                         train_samples  eval_samples division  \\\n",
       "submission_label                                                                \n",
       "LBNL-1/corigpu_n64_pt1.6.0                      121216         15158   closed   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed         121266         15158   closed   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open         121266         15158     open   \n",
       "\n",
       "                                        submission_long_label  \\\n",
       "submission_label                                                \n",
       "LBNL-1/corigpu_n64_pt1.6.0                        Cori-GPU-64   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed             ABCI-1024   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open             ABCI-1024   \n",
       "\n",
       "                                        submission_epochs_label  \\\n",
       "submission_label                                                  \n",
       "LBNL-1/corigpu_n64_pt1.6.0                        Cori\\n-GPU-64   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed               ABCI-1024   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open             ABCI\\n-1024   \n",
       "\n",
       "                                         eval_time/run_time  \\\n",
       "submission_label                                              \n",
       "LBNL-1/corigpu_n64_pt1.6.0                         0.362732   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed            0.040034   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open            0.013057   \n",
       "\n",
       "                                         run_avg_epoch_sample_throughput.mean  \\\n",
       "submission_label                                                                \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                       13636.646533   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                         328452.837177   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                         341240.183868   \n",
       "\n",
       "                                         run_avg_epoch_sample_throughput.std  \\\n",
       "submission_label                                                               \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                        307.016017   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                           954.164474   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                          8540.870875   \n",
       "\n",
       "                                         run_avg_epoch_samples.mean  \\\n",
       "submission_label                                                      \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                 189932.0   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                    130108.0   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                    122953.0   \n",
       "\n",
       "                                         run_avg_epoch_samples.std  \\\n",
       "submission_label                                                     \n",
       "LBNL-1/corigpu_n64_pt1.6.0                              874.685658   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                   0.000000   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                 639.126748   \n",
       "\n",
       "                                         run_epoch_throughput.mean  \\\n",
       "submission_label                                                     \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                0.071793   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                   2.524463   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                   2.775585   \n",
       "\n",
       "                                         run_epoch_throughput.std  \\\n",
       "submission_label                                                    \n",
       "LBNL-1/corigpu_n64_pt1.6.0                               0.001919   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                  0.007334   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                  0.056186   \n",
       "\n",
       "                                         run_epoch_time.mean  \\\n",
       "submission_label                                               \n",
       "LBNL-1/corigpu_n64_pt1.6.0                         13.928849   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed             0.396124   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open             0.360284   \n",
       "\n",
       "                                         run_epoch_time.std  \\\n",
       "submission_label                                              \n",
       "LBNL-1/corigpu_n64_pt1.6.0                         0.362650   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed            0.001147   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open            0.007149   \n",
       "\n",
       "                                         run_eval_throughput.mean  \\\n",
       "submission_label                                                    \n",
       "LBNL-1/corigpu_n64_pt1.6.0                           13605.274039   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed             452842.370981   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open             281674.493324   \n",
       "\n",
       "                                         run_eval_throughput.std  \\\n",
       "submission_label                                                   \n",
       "LBNL-1/corigpu_n64_pt1.6.0                            704.124769   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed               427.222282   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open             53363.772466   \n",
       "\n",
       "                                         run_eval_time.mean  \\\n",
       "submission_label                                              \n",
       "LBNL-1/corigpu_n64_pt1.6.0                        50.524406   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed            0.468622   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open            0.136906   \n",
       "\n",
       "                                         run_eval_time.std  \\\n",
       "submission_label                                             \n",
       "LBNL-1/corigpu_n64_pt1.6.0                        3.044168   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed           0.000442   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open           0.033568   \n",
       "\n",
       "                                         run_num_evals.mean  \\\n",
       "submission_label                                              \n",
       "LBNL-1/corigpu_n64_pt1.6.0                        45.333333   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed           14.000000   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open            2.666667   \n",
       "\n",
       "                                         run_num_evals.std  \\\n",
       "submission_label                                             \n",
       "LBNL-1/corigpu_n64_pt1.6.0                        0.577350   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed           0.000000   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open           1.154701   \n",
       "\n",
       "                                         run_staging_throughput.mean  \\\n",
       "submission_label                                                       \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                       NaN   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                 62072.354483   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                 69521.318153   \n",
       "\n",
       "                                         run_staging_throughput.std  \\\n",
       "submission_label                                                      \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                      NaN   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                  180.453773   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                 2846.825395   \n",
       "\n",
       "                                         run_staging_time.mean  \\\n",
       "submission_label                                                 \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                 NaN   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed               2.197822   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open               1.962333   \n",
       "\n",
       "                                         run_staging_time.std  \\\n",
       "submission_label                                                \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                NaN   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed              0.006371   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open              0.077194   \n",
       "\n",
       "                                         run_stop_epoch.mean  \\\n",
       "submission_label                                               \n",
       "LBNL-1/corigpu_n64_pt1.6.0                         10.000000   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed            24.000000   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open            23.666667   \n",
       "\n",
       "                                         run_stop_epoch.std  run_time.mean  \\\n",
       "submission_label                                                             \n",
       "LBNL-1/corigpu_n64_pt1.6.0                         0.000000     139.288611   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed            0.000000      11.705717   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open            1.154701      10.485394   \n",
       "\n",
       "                                         run_time.std  \\\n",
       "submission_label                                        \n",
       "LBNL-1/corigpu_n64_pt1.6.0                   3.626479   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed      0.021555   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open      0.228172   \n",
       "\n",
       "                                         run_train_throughput.mean  \\\n",
       "submission_label                                                     \n",
       "LBNL-1/corigpu_n64_pt1.6.0                            13655.973840   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed              322003.905651   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open              342041.745864   \n",
       "\n",
       "                                         run_train_throughput.std  \\\n",
       "submission_label                                                    \n",
       "LBNL-1/corigpu_n64_pt1.6.0                             496.046978   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                993.958276   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open               8021.315726   \n",
       "\n",
       "                                         run_train_time.mean  \\\n",
       "submission_label                                               \n",
       "LBNL-1/corigpu_n64_pt1.6.0                         88.764083   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed             9.038350   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open             8.385078   \n",
       "\n",
       "                                         run_train_time.std  \\\n",
       "submission_label                                              \n",
       "LBNL-1/corigpu_n64_pt1.6.0                         3.111298   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed            0.027814   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open            0.246814   \n",
       "\n",
       "                                         staging_time/run_time  \\\n",
       "submission_label                                                 \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                 NaN   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed               0.187756   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open               0.187149   \n",
       "\n",
       "                                         train_time/run_time  \n",
       "submission_label                                              \n",
       "LBNL-1/corigpu_n64_pt1.6.0                          0.637267  \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed             0.772131  \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open             0.799691  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "def throughput_aware_mean(run_metric):\n",
    "    if not run_metric.name.endswith('throughput'):\n",
    "        return run_metric.mean()\n",
    "    else:\n",
    "        return 1./(1./run_metric).mean()\n",
    "\n",
    "def throughput_aware_std(run_metric):\n",
    "    if not run_metric.name.endswith('throughput'):\n",
    "        return run_metric.std()\n",
    "    else:\n",
    "        mean = throughput_aware_mean(run_metric)\n",
    "        return 1./(1./mean - (1./run_metric).std()) - mean\n",
    "\n",
    "submissions_stats_df_run_mean = submissions_stats_df['runs'].apply(lambda expt: pd.DataFrame.from_dict(expt).apply(throughput_aware_mean)).add_suffix('.mean')\n",
    "submissions_stats_df_run_mean.index = submissions_stats_df.index\n",
    "submissions_stats_df_run_std = submissions_stats_df['runs'].apply(lambda expt: pd.DataFrame.from_dict(expt).apply(throughput_aware_std)).add_suffix('.std')\n",
    "submissions_stats_df_run_std.index = submissions_stats_df.index\n",
    "submissions_stats_df_run = submissions_stats_df_run_mean.join(submissions_stats_df_run_std)\n",
    "\n",
    "for c in submissions_stats_df_run.columns: # transform units from seconds to minutes\n",
    "    if 'throughput' in c:\n",
    "        submissions_stats_df_run[c] *= 60.\n",
    "    elif 'time' in c:\n",
    "        submissions_stats_df_run[c] /= 60.\n",
    "\n",
    "submissions_stats_df_run['staging_time/run_time'] = submissions_stats_df_run.apply(lambda row: row['run_staging_time.mean']/row['run_time.mean'], axis=1)\n",
    "submissions_stats_df_run['train_time/run_time']   = submissions_stats_df_run.apply(lambda row: row['run_train_time.mean']/row['run_time.mean'], axis=1)\n",
    "submissions_stats_df_run['eval_time/run_time']    = submissions_stats_df_run.apply(lambda row: row['run_eval_time.mean']/row['run_time.mean'], axis=1)\n",
    "\n",
    "submissions_stats_df_run = submissions_stats_df_run[sorted(submissions_stats_df_run.columns)]\n",
    "submissions_stats_df_run.index = submissions_stats_df.index\n",
    "submissions_stats_df_run\n",
    "\n",
    "submissions_stats_df_run_vis = submissions_stats_df.drop(columns=['submission_short_label', 'runs', 'opt_opts', 'system']).join(submissions_stats_df_run).set_index('submission_label').sort_values(['division', 'global_batch_size'])\n",
    "display(submissions_stats_df_run_vis)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "# if benchmark.round == 'hpc_results_v0.7':\n",
    "#     if benchmark.name == 'cosmoflow':\n",
    "#\n",
    "#         # Run-to-run variation\n",
    "#         print((submissions_stats_df_run_vis[submissions_stats_df_run_vis['global_batch_size'] <= 1024]['run_time.std']/submissions_stats_df_run_vis[submissions_stats_df_run_vis['global_batch_size'] <= 1024]['run_time.mean']).mean()) #.to_latex(float_format=\"{:0.2f}\".format)) #, 'staging_time/run_time', 'train_time/run_time', 'eval_time/run_time']] \n",
    "#\n",
    "#         # Throughput scaling\n",
    "#         submissions_stats_df_run_vis.loc['CSCS-1/daint_gpu_n256_tf2.2.0',['run_train_throughput.mean', 'run_eval_throughput.mean',  'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]/2./submissions_stats_df_run_vis.loc['CSCS-2/daint_gpu_n128_tf2.2.0',['run_train_throughput.mean', 'run_eval_throughput.mean', 'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]\n",
    "#         submissions_stats_df_run_vis.loc['LBNL-2/coriknl_n1024_tf1.15.2',['run_train_throughput.mean', 'run_eval_throughput.mean',  'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]/2./submissions_stats_df_run_vis.loc['LBNL-1/coriknl_n512_tf1.15.2',['run_train_throughput.mean', 'run_eval_throughput.mean', 'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]\n",
    "#         submissions_stats_df_run_vis.loc['Fujitsu-1/abci_2048xV100_tensorflow_open',['run_train_throughput.mean', 'run_eval_throughput.mean',  'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]/4./submissions_stats_df_run_vis.loc['Fujitsu-4/abci_512xV100_tensorflow_closed',['run_train_throughput.mean', 'run_eval_throughput.mean', 'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]\n",
    "#         submissions_stats_df_run_vis.loc['Fujitsu-3/fugaku_8192xA64FX_tensorflow_closed',['run_train_throughput.mean', 'run_eval_throughput.mean',  'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]/16./submissions_stats_df_run_vis.loc['RIKEN-1/fugaku_512xA64FX_tensorflow_closed',['run_train_throughput.mean', 'run_eval_throughput.mean', 'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]\n",
    "#         submissions_stats_df_run_vis.loc['Fujitsu-2/fugaku_16384xA64FX_tensorflow_open',['run_train_throughput.mean', 'run_eval_throughput.mean',  'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]/8./4./submissions_stats_df_run_vis.loc['RIKEN-1/fugaku_512xA64FX_tensorflow_closed',['run_train_throughput.mean', 'run_eval_throughput.mean', 'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]\n",
    "#         submissions_stats_df_run_vis.loc['LBNL-3/corigpu_n64_tf1.15.0',['run_train_throughput.mean', 'run_eval_throughput.mean',  'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]*8./submissions_stats_df_run_vis.loc['Fujitsu-4/abci_512xV100_tensorflow_closed',['run_train_throughput.mean', 'run_eval_throughput.mean', 'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]\n",
    "#         submissions_stats_df_run_vis.loc['NCSA-1/halv100_n16_tf1.15.0',['run_train_throughput.mean', 'run_eval_throughput.mean',  'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]*8./submissions_stats_df_run_vis.loc['Fujitsu-4/abci_512xV100_tensorflow_closed',['run_train_throughput.mean', 'run_eval_throughput.mean', 'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]\n",
    "#         submissions_stats_df_run_vis.loc['Fujitsu-4/abci_512xV100_tensorflow_closed',['run_train_throughput.mean', 'run_eval_throughput.mean',  'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]/8./submissions_stats_df_run_vis.loc['LBNL-3/corigpu_n64_tf1.15.0',['run_train_throughput.mean', 'run_eval_throughput.mean', 'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]\n",
    "#         submissions_stats_df_run_vis.loc['Fujitsu-4/abci_512xV100_tensorflow_closed',['run_train_throughput.mean', 'run_eval_throughput.mean',  'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]/8./submissions_stats_df_run_vis.loc['NCSA-1/halv100_n16_tf1.15.0',['run_train_throughput.mean', 'run_eval_throughput.mean', 'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]\n",
    "#         submissions_stats_df_run_vis.loc['NCSA-1/halv100_n16_tf1.15.0',['run_train_throughput.mean', 'run_eval_throughput.mean',  'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]/submissions_stats_df_run_vis.loc['LBNL-3/corigpu_n64_tf1.15.0',['run_train_throughput.mean', 'run_eval_throughput.mean', 'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]\n",
    "#         submissions_stats_df_run_vis.loc['LBNL-1/corigpu_n64_pt1.6.0',['run_train_throughput.mean', 'run_eval_throughput.mean',  'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]*16./submissions_stats_df_run_vis.loc['Fujitsu-1/abci_1024xV100_pytorch_closed',['run_train_throughput.mean', 'run_eval_throughput.mean', 'run_avg_epoch_sample_throughput.mean', 'run_epoch_throughput.mean']]\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Data staging analysis: Relative time spent in staging, evaluation and training and T_staging/T_epoch-ratio"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>staging_time/run_time</th>\n",
       "      <th>eval_time/run_time</th>\n",
       "      <th>train_time/run_time</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>submission_label</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>LBNL-1/corigpu_n64_pt1.6.0</th>\n",
       "      <td>NaN</td>\n",
       "      <td>0.362732</td>\n",
       "      <td>0.637267</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fujitsu-1/abci_1024xV100_pytorch_closed</th>\n",
       "      <td>0.187756</td>\n",
       "      <td>0.040034</td>\n",
       "      <td>0.772131</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fujitsu-2/abci_1024xV100_fjpytorch_open</th>\n",
       "      <td>0.187149</td>\n",
       "      <td>0.013057</td>\n",
       "      <td>0.799691</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                         staging_time/run_time  eval_time/run_time  train_time/run_time\n",
       "submission_label                                                                                       \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                 NaN            0.362732             0.637267\n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed               0.187756            0.040034             0.772131\n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open               0.187149            0.013057             0.799691"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x99.6923 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "if benchmark.round == 'hpc_results_v0.7':\n",
    "    if benchmark.name == 'cosmoflow':\n",
    "        submissions_stats_df_run_vis_cosmoflow = submissions_stats_df_run_vis\n",
    "        submissions_stats_df_run_vis_cosmoflow = submissions_stats_df_run_vis_cosmoflow.reindex([\n",
    "            'CSCS-2/daint_gpu_n128_tf2.2.0',\n",
    "            'CSCS-1/daint_gpu_n256_tf2.2.0', \n",
    "            'Fujitsu-4/abci_512xV100_tensorflow_closed',\n",
    "            'Fujitsu-1/abci_2048xV100_tensorflow_open',\n",
    "            'RIKEN-1/fugaku_512xA64FX_tensorflow_closed',\n",
    "            'Fujitsu-3/fugaku_8192xA64FX_tensorflow_closed',\n",
    "            'Fujitsu-2/fugaku_16384xA64FX_tensorflow_open',\n",
    "            'LBNL-3/corigpu_n64_tf1.15.0',\n",
    "            'LBNL-1/coriknl_n512_tf1.15.2',\n",
    "            'LBNL-2/coriknl_n1024_tf1.15.2',\n",
    "            'NCSA-1/halv100_n16_tf1.15.0', \n",
    "            'TACC-1/frontera_rtx_n16_tf1.15.2_lbs1'])\n",
    "        submissions_stats_df_run_vis_all = submissions_stats_df_run_vis_cosmoflow\n",
    "    else:\n",
    "        submissions_stats_df_run_vis_deepcam = submissions_stats_df_run_vis.reindex([\n",
    "            'LBNL-1/corigpu_n64_pt1.6.0', \n",
    "            'Fujitsu-1/abci_1024xV100_pytorch_closed',\n",
    "            'Fujitsu-2/abci_1024xV100_fjpytorch_open'])\n",
    "        submissions_stats_df_run_vis_all = submissions_stats_df_run_vis_deepcam\n",
    "        if 'submissions_stats_df_run_vis_cosmoflow' in locals():\n",
    "            submissions_stats_df_run_vis_all = pd.concat([submissions_stats_df_run_vis_cosmoflow, submissions_stats_df_run_vis_deepcam])\n",
    "    display(submissions_stats_df_run_vis_all[['staging_time/run_time', 'eval_time/run_time', 'train_time/run_time']])\n",
    "\n",
    "    def generate_relative_staging_eval_training_times_plot():\n",
    "        def create_relative_time_plot(df, ax, split=None):\n",
    "            ind = np.arange(len(df.index), 0, -1) + 0.0\n",
    "            if split is not None:\n",
    "                ind[:split] += .66\n",
    "\n",
    "            ax.barh(ind, df['eval_time/run_time'] + df['train_time/run_time'] + df['staging_time/run_time'].apply(lambda val: 0. if pd.isnull(val) else val), label='train')\n",
    "            ax.barh(ind, df['eval_time/run_time'] + df['staging_time/run_time'].apply(lambda val: 0. if pd.isnull(val) else val), label='eval')\n",
    "            ax.barh(ind, df['staging_time/run_time'].apply(lambda val: 0. if pd.isnull(val) else val), label='staging')\n",
    "\n",
    "            ax.set_yticks(ind)\n",
    "            ax.set_yticklabels(df.apply(lambda row: (r'$\\bigstar$' if row['division'] == 'open' else '') + row['submission_long_label'].replace('\\n',''), axis=1))\n",
    "            ax.xaxis.set_major_locator(plt.MultipleLocator(0.1))\n",
    "            ax.set_xticks(np.linspace(0,1.,41), minor=True)\n",
    "            ax.set_xlim([0,1])\n",
    "            ax.grid(True, which='major', axis='x', alpha=1.)\n",
    "            ax.grid(True, which='minor', axis='x', alpha=0.5)\n",
    "            ax.set_ylim(0.4,df.shape[0] + 0.8 + (0. if split is None else 0.5))\n",
    "\n",
    "        if 'submissions_stats_df_run_vis_cosmoflow' in locals() and 'submissions_stats_df_run_vis_deepcam' in locals():\n",
    "            fig, ax = plt.subplots(figsize=(10,4.5))\n",
    "            create_relative_time_plot(submissions_stats_df_run_vis_all, ax, split=int(-submissions_stats_df_run_vis_deepcam.shape[0]))\n",
    "        else:\n",
    "            fig, ax = plt.subplots(figsize=(10,4.5 if benchmark.name == 'cosmoflow' else 4.5/13*4))\n",
    "            create_relative_time_plot(submissions_stats_df_run_vis_all, ax, split=None)        \n",
    "        fig.tight_layout()\n",
    "#     generate_relative_staging_eval_training_times_plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>run_time.mean</th>\n",
       "      <th>run_time.std</th>\n",
       "      <th>run_staging_time.mean</th>\n",
       "      <th>run_staging_time.std</th>\n",
       "      <th>run_epoch_time.mean</th>\n",
       "      <th>run_epoch_time.std</th>\n",
       "      <th>run_staging_epoch_time_ratio</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>submission_label</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Fujitsu-1/abci_1024xV100_pytorch_closed</th>\n",
       "      <td>702.343000</td>\n",
       "      <td>1.293315</td>\n",
       "      <td>131.869333</td>\n",
       "      <td>0.382253</td>\n",
       "      <td>23.767431</td>\n",
       "      <td>0.068845</td>\n",
       "      <td>5.548321</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fujitsu-2/abci_1024xV100_fjpytorch_open</th>\n",
       "      <td>629.123667</td>\n",
       "      <td>13.690344</td>\n",
       "      <td>117.740000</td>\n",
       "      <td>4.631668</td>\n",
       "      <td>21.617065</td>\n",
       "      <td>0.428912</td>\n",
       "      <td>5.446623</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                         run_time.mean  run_time.std  \\\n",
       "submission_label                                                       \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed     702.343000      1.293315   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open     629.123667     13.690344   \n",
       "\n",
       "                                         run_staging_time.mean  \\\n",
       "submission_label                                                 \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed             131.869333   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open             117.740000   \n",
       "\n",
       "                                         run_staging_time.std  \\\n",
       "submission_label                                                \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed              0.382253   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open              4.631668   \n",
       "\n",
       "                                         run_epoch_time.mean  \\\n",
       "submission_label                                               \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed            23.767431   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open            21.617065   \n",
       "\n",
       "                                         run_epoch_time.std  \\\n",
       "submission_label                                              \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed            0.068845   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open            0.428912   \n",
       "\n",
       "                                         run_staging_epoch_time_ratio  \n",
       "submission_label                                                       \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                      5.548321  \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                      5.446623  "
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Staging to epoch time ratio\n",
    "if benchmark.round == 'hpc_results_v0.7':  # show staging time and relation to compute\n",
    "    submissions_stats_df_run_vis_times = submissions_stats_df_run_vis[ ['run_time.mean', 'run_time.std'] + \n",
    "                                                                       ['run_staging_time.mean', 'run_staging_time.std'] + \n",
    "                                                                       ['run_epoch_time.mean', 'run_epoch_time.std'] ].applymap(lambda s: s*60.)\n",
    "    submissions_stats_df_run_vis_times['run_staging_epoch_time_ratio'] = \\\n",
    "                                                    submissions_stats_df_run_vis.apply(lambda s: s['run_staging_time.mean']/s['run_epoch_time.mean'], axis=1)\n",
    "submissions_stats_df_run_vis_times.dropna()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Comparing systems with submissions to both benchmarks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>global_batch_size</th>\n",
       "      <th>opt_name</th>\n",
       "      <th>train_samples</th>\n",
       "      <th>eval_samples</th>\n",
       "      <th>division</th>\n",
       "      <th>submission_long_label</th>\n",
       "      <th>submission_epochs_label</th>\n",
       "      <th>eval_time/run_time</th>\n",
       "      <th>run_avg_epoch_sample_throughput.mean</th>\n",
       "      <th>run_avg_epoch_sample_throughput.std</th>\n",
       "      <th>run_avg_epoch_samples.mean</th>\n",
       "      <th>run_avg_epoch_samples.std</th>\n",
       "      <th>run_epoch_throughput.mean</th>\n",
       "      <th>run_epoch_throughput.std</th>\n",
       "      <th>run_epoch_time.mean</th>\n",
       "      <th>run_epoch_time.std</th>\n",
       "      <th>run_eval_throughput.mean</th>\n",
       "      <th>run_eval_throughput.std</th>\n",
       "      <th>run_eval_time.mean</th>\n",
       "      <th>run_eval_time.std</th>\n",
       "      <th>run_num_evals.mean</th>\n",
       "      <th>run_num_evals.std</th>\n",
       "      <th>run_staging_throughput.mean</th>\n",
       "      <th>run_staging_throughput.std</th>\n",
       "      <th>run_staging_time.mean</th>\n",
       "      <th>run_staging_time.std</th>\n",
       "      <th>run_stop_epoch.mean</th>\n",
       "      <th>run_stop_epoch.std</th>\n",
       "      <th>run_time.mean</th>\n",
       "      <th>run_time.std</th>\n",
       "      <th>run_train_throughput.mean</th>\n",
       "      <th>run_train_throughput.std</th>\n",
       "      <th>run_train_time.mean</th>\n",
       "      <th>run_train_time.std</th>\n",
       "      <th>staging_time/run_time</th>\n",
       "      <th>train_time/run_time</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>submission_label</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>LBNL-1/corigpu_n64_pt1.6.0</th>\n",
       "      <td>128</td>\n",
       "      <td>LAMB</td>\n",
       "      <td>121216</td>\n",
       "      <td>15158</td>\n",
       "      <td>closed</td>\n",
       "      <td>Cori-GPU-64</td>\n",
       "      <td>Cori\\n-GPU-64</td>\n",
       "      <td>0.362732</td>\n",
       "      <td>13636.646533</td>\n",
       "      <td>307.016017</td>\n",
       "      <td>189932.0</td>\n",
       "      <td>874.685658</td>\n",
       "      <td>0.071793</td>\n",
       "      <td>0.001919</td>\n",
       "      <td>13.928849</td>\n",
       "      <td>0.362650</td>\n",
       "      <td>13605.274039</td>\n",
       "      <td>704.124769</td>\n",
       "      <td>50.524406</td>\n",
       "      <td>3.044168</td>\n",
       "      <td>45.333333</td>\n",
       "      <td>0.577350</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>10.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>139.288611</td>\n",
       "      <td>3.626479</td>\n",
       "      <td>13655.973840</td>\n",
       "      <td>496.046978</td>\n",
       "      <td>88.764083</td>\n",
       "      <td>3.111298</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.637267</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fujitsu-1/abci_1024xV100_pytorch_closed</th>\n",
       "      <td>2048</td>\n",
       "      <td>LAMB</td>\n",
       "      <td>121266</td>\n",
       "      <td>15158</td>\n",
       "      <td>closed</td>\n",
       "      <td>ABCI-1024</td>\n",
       "      <td>ABCI-1024</td>\n",
       "      <td>0.040034</td>\n",
       "      <td>328452.837177</td>\n",
       "      <td>954.164474</td>\n",
       "      <td>130108.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>2.524463</td>\n",
       "      <td>0.007334</td>\n",
       "      <td>0.396124</td>\n",
       "      <td>0.001147</td>\n",
       "      <td>452842.370981</td>\n",
       "      <td>427.222282</td>\n",
       "      <td>0.468622</td>\n",
       "      <td>0.000442</td>\n",
       "      <td>14.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>62072.354483</td>\n",
       "      <td>180.453773</td>\n",
       "      <td>2.197822</td>\n",
       "      <td>0.006371</td>\n",
       "      <td>24.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>11.705717</td>\n",
       "      <td>0.021555</td>\n",
       "      <td>322003.905651</td>\n",
       "      <td>993.958276</td>\n",
       "      <td>9.038350</td>\n",
       "      <td>0.027814</td>\n",
       "      <td>0.187756</td>\n",
       "      <td>0.772131</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fujitsu-2/abci_1024xV100_fjpytorch_open</th>\n",
       "      <td>2048</td>\n",
       "      <td>LAMB</td>\n",
       "      <td>121266</td>\n",
       "      <td>15158</td>\n",
       "      <td>open</td>\n",
       "      <td>ABCI-1024</td>\n",
       "      <td>ABCI\\n-1024</td>\n",
       "      <td>0.013057</td>\n",
       "      <td>341240.183868</td>\n",
       "      <td>8540.870875</td>\n",
       "      <td>122953.0</td>\n",
       "      <td>639.126748</td>\n",
       "      <td>2.775585</td>\n",
       "      <td>0.056186</td>\n",
       "      <td>0.360284</td>\n",
       "      <td>0.007149</td>\n",
       "      <td>281674.493324</td>\n",
       "      <td>53363.772466</td>\n",
       "      <td>0.136906</td>\n",
       "      <td>0.033568</td>\n",
       "      <td>2.666667</td>\n",
       "      <td>1.154701</td>\n",
       "      <td>69521.318153</td>\n",
       "      <td>2846.825395</td>\n",
       "      <td>1.962333</td>\n",
       "      <td>0.077194</td>\n",
       "      <td>23.666667</td>\n",
       "      <td>1.154701</td>\n",
       "      <td>10.485394</td>\n",
       "      <td>0.228172</td>\n",
       "      <td>342041.745864</td>\n",
       "      <td>8021.315726</td>\n",
       "      <td>8.385078</td>\n",
       "      <td>0.246814</td>\n",
       "      <td>0.187149</td>\n",
       "      <td>0.799691</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                         global_batch_size opt_name  \\\n",
       "submission_label                                                      \n",
       "LBNL-1/corigpu_n64_pt1.6.0                             128     LAMB   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed               2048     LAMB   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open               2048     LAMB   \n",
       "\n",
       "                                         train_samples  eval_samples division  \\\n",
       "submission_label                                                                \n",
       "LBNL-1/corigpu_n64_pt1.6.0                      121216         15158   closed   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed         121266         15158   closed   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open         121266         15158     open   \n",
       "\n",
       "                                        submission_long_label  \\\n",
       "submission_label                                                \n",
       "LBNL-1/corigpu_n64_pt1.6.0                        Cori-GPU-64   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed             ABCI-1024   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open             ABCI-1024   \n",
       "\n",
       "                                        submission_epochs_label  \\\n",
       "submission_label                                                  \n",
       "LBNL-1/corigpu_n64_pt1.6.0                        Cori\\n-GPU-64   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed               ABCI-1024   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open             ABCI\\n-1024   \n",
       "\n",
       "                                         eval_time/run_time  \\\n",
       "submission_label                                              \n",
       "LBNL-1/corigpu_n64_pt1.6.0                         0.362732   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed            0.040034   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open            0.013057   \n",
       "\n",
       "                                         run_avg_epoch_sample_throughput.mean  \\\n",
       "submission_label                                                                \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                       13636.646533   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                         328452.837177   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                         341240.183868   \n",
       "\n",
       "                                         run_avg_epoch_sample_throughput.std  \\\n",
       "submission_label                                                               \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                        307.016017   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                           954.164474   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                          8540.870875   \n",
       "\n",
       "                                         run_avg_epoch_samples.mean  \\\n",
       "submission_label                                                      \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                 189932.0   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                    130108.0   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                    122953.0   \n",
       "\n",
       "                                         run_avg_epoch_samples.std  \\\n",
       "submission_label                                                     \n",
       "LBNL-1/corigpu_n64_pt1.6.0                              874.685658   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                   0.000000   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                 639.126748   \n",
       "\n",
       "                                         run_epoch_throughput.mean  \\\n",
       "submission_label                                                     \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                0.071793   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                   2.524463   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                   2.775585   \n",
       "\n",
       "                                         run_epoch_throughput.std  \\\n",
       "submission_label                                                    \n",
       "LBNL-1/corigpu_n64_pt1.6.0                               0.001919   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                  0.007334   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                  0.056186   \n",
       "\n",
       "                                         run_epoch_time.mean  \\\n",
       "submission_label                                               \n",
       "LBNL-1/corigpu_n64_pt1.6.0                         13.928849   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed             0.396124   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open             0.360284   \n",
       "\n",
       "                                         run_epoch_time.std  \\\n",
       "submission_label                                              \n",
       "LBNL-1/corigpu_n64_pt1.6.0                         0.362650   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed            0.001147   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open            0.007149   \n",
       "\n",
       "                                         run_eval_throughput.mean  \\\n",
       "submission_label                                                    \n",
       "LBNL-1/corigpu_n64_pt1.6.0                           13605.274039   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed             452842.370981   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open             281674.493324   \n",
       "\n",
       "                                         run_eval_throughput.std  \\\n",
       "submission_label                                                   \n",
       "LBNL-1/corigpu_n64_pt1.6.0                            704.124769   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed               427.222282   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open             53363.772466   \n",
       "\n",
       "                                         run_eval_time.mean  \\\n",
       "submission_label                                              \n",
       "LBNL-1/corigpu_n64_pt1.6.0                        50.524406   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed            0.468622   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open            0.136906   \n",
       "\n",
       "                                         run_eval_time.std  \\\n",
       "submission_label                                             \n",
       "LBNL-1/corigpu_n64_pt1.6.0                        3.044168   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed           0.000442   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open           0.033568   \n",
       "\n",
       "                                         run_num_evals.mean  \\\n",
       "submission_label                                              \n",
       "LBNL-1/corigpu_n64_pt1.6.0                        45.333333   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed           14.000000   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open            2.666667   \n",
       "\n",
       "                                         run_num_evals.std  \\\n",
       "submission_label                                             \n",
       "LBNL-1/corigpu_n64_pt1.6.0                        0.577350   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed           0.000000   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open           1.154701   \n",
       "\n",
       "                                         run_staging_throughput.mean  \\\n",
       "submission_label                                                       \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                       NaN   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                 62072.354483   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                 69521.318153   \n",
       "\n",
       "                                         run_staging_throughput.std  \\\n",
       "submission_label                                                      \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                      NaN   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                  180.453773   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                 2846.825395   \n",
       "\n",
       "                                         run_staging_time.mean  \\\n",
       "submission_label                                                 \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                 NaN   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed               2.197822   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open               1.962333   \n",
       "\n",
       "                                         run_staging_time.std  \\\n",
       "submission_label                                                \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                NaN   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed              0.006371   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open              0.077194   \n",
       "\n",
       "                                         run_stop_epoch.mean  \\\n",
       "submission_label                                               \n",
       "LBNL-1/corigpu_n64_pt1.6.0                         10.000000   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed            24.000000   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open            23.666667   \n",
       "\n",
       "                                         run_stop_epoch.std  run_time.mean  \\\n",
       "submission_label                                                             \n",
       "LBNL-1/corigpu_n64_pt1.6.0                         0.000000     139.288611   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed            0.000000      11.705717   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open            1.154701      10.485394   \n",
       "\n",
       "                                         run_time.std  \\\n",
       "submission_label                                        \n",
       "LBNL-1/corigpu_n64_pt1.6.0                   3.626479   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed      0.021555   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open      0.228172   \n",
       "\n",
       "                                         run_train_throughput.mean  \\\n",
       "submission_label                                                     \n",
       "LBNL-1/corigpu_n64_pt1.6.0                            13655.973840   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed              322003.905651   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open              342041.745864   \n",
       "\n",
       "                                         run_train_throughput.std  \\\n",
       "submission_label                                                    \n",
       "LBNL-1/corigpu_n64_pt1.6.0                             496.046978   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                993.958276   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open               8021.315726   \n",
       "\n",
       "                                         run_train_time.mean  \\\n",
       "submission_label                                               \n",
       "LBNL-1/corigpu_n64_pt1.6.0                         88.764083   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed             9.038350   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open             8.385078   \n",
       "\n",
       "                                         run_train_time.std  \\\n",
       "submission_label                                              \n",
       "LBNL-1/corigpu_n64_pt1.6.0                         3.111298   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed            0.027814   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open            0.246814   \n",
       "\n",
       "                                         staging_time/run_time  \\\n",
       "submission_label                                                 \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                 NaN   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed               0.187756   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open               0.187149   \n",
       "\n",
       "                                         train_time/run_time  \n",
       "submission_label                                              \n",
       "LBNL-1/corigpu_n64_pt1.6.0                          0.637267  \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed             0.772131  \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open             0.799691  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "if benchmark.round == 'hpc_results_v0.7':\n",
    "    if benchmark.name == 'cosmoflow':\n",
    "        submissions_to_show = ['LBNL-3/corigpu_n64_tf1.15.0',\n",
    "                               'Fujitsu-4/abci_512xV100_tensorflow_closed',\n",
    "                               'Fujitsu-1/abci_2048xV100_tensorflow_open']\n",
    "    else:\n",
    "        submissions_to_show = ['LBNL-1/corigpu_n64_pt1.6.0', \n",
    "                               'Fujitsu-1/abci_1024xV100_pytorch_closed',\n",
    "                               'Fujitsu-2/abci_1024xV100_fjpytorch_open']\n",
    "\n",
    "    display(submissions_stats_df_run_vis.loc[submissions_to_show,:])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Optimizer convergence"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>train_samples</th>\n",
       "      <th>eval_samples</th>\n",
       "      <th>division</th>\n",
       "      <th>submission_long_label</th>\n",
       "      <th>submission_epochs_label</th>\n",
       "      <th>run_stop_epoch.mean</th>\n",
       "      <th>run_stop_epoch.std</th>\n",
       "      <th>global_batch_size</th>\n",
       "      <th>opt_name</th>\n",
       "      <th>opt_epsilon</th>\n",
       "      <th>opt_base_learning_rate</th>\n",
       "      <th>opt_learning_rate_warmup_steps</th>\n",
       "      <th>opt_learning_rate_warmup_factor</th>\n",
       "      <th>opt_weight_decay</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>submission_label</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>LBNL-1/corigpu_n64_pt1.6.0</th>\n",
       "      <td>121216</td>\n",
       "      <td>15158</td>\n",
       "      <td>closed</td>\n",
       "      <td>Cori-GPU-64</td>\n",
       "      <td>Cori\\n-GPU-64</td>\n",
       "      <td>10.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>128</td>\n",
       "      <td>LAMB</td>\n",
       "      <td>1.000000e-08</td>\n",
       "      <td>0.0010</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fujitsu-1/abci_1024xV100_pytorch_closed</th>\n",
       "      <td>121266</td>\n",
       "      <td>15158</td>\n",
       "      <td>closed</td>\n",
       "      <td>ABCI-1024</td>\n",
       "      <td>ABCI-1024</td>\n",
       "      <td>24.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>2048</td>\n",
       "      <td>LAMB</td>\n",
       "      <td>1.000000e-08</td>\n",
       "      <td>0.0055</td>\n",
       "      <td>400.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fujitsu-2/abci_1024xV100_fjpytorch_open</th>\n",
       "      <td>121266</td>\n",
       "      <td>15158</td>\n",
       "      <td>open</td>\n",
       "      <td>ABCI-1024</td>\n",
       "      <td>ABCI\\n-1024</td>\n",
       "      <td>23.666667</td>\n",
       "      <td>1.154701</td>\n",
       "      <td>2048</td>\n",
       "      <td>LAMB</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                         train_samples  eval_samples division  \\\n",
       "submission_label                                                                \n",
       "LBNL-1/corigpu_n64_pt1.6.0                      121216         15158   closed   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed         121266         15158   closed   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open         121266         15158     open   \n",
       "\n",
       "                                        submission_long_label  \\\n",
       "submission_label                                                \n",
       "LBNL-1/corigpu_n64_pt1.6.0                        Cori-GPU-64   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed             ABCI-1024   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open             ABCI-1024   \n",
       "\n",
       "                                        submission_epochs_label  \\\n",
       "submission_label                                                  \n",
       "LBNL-1/corigpu_n64_pt1.6.0                        Cori\\n-GPU-64   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed               ABCI-1024   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open             ABCI\\n-1024   \n",
       "\n",
       "                                         run_stop_epoch.mean  \\\n",
       "submission_label                                               \n",
       "LBNL-1/corigpu_n64_pt1.6.0                         10.000000   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed            24.000000   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open            23.666667   \n",
       "\n",
       "                                         run_stop_epoch.std  \\\n",
       "submission_label                                              \n",
       "LBNL-1/corigpu_n64_pt1.6.0                         0.000000   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed            0.000000   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open            1.154701   \n",
       "\n",
       "                                         global_batch_size opt_name  \\\n",
       "submission_label                                                      \n",
       "LBNL-1/corigpu_n64_pt1.6.0                             128     LAMB   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed               2048     LAMB   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open               2048     LAMB   \n",
       "\n",
       "                                          opt_epsilon  opt_base_learning_rate  \\\n",
       "submission_label                                                                \n",
       "LBNL-1/corigpu_n64_pt1.6.0               1.000000e-08                  0.0010   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed  1.000000e-08                  0.0055   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open           NaN                     NaN   \n",
       "\n",
       "                                         opt_learning_rate_warmup_steps  \\\n",
       "submission_label                                                          \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                          0.0   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                           400.0   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                             NaN   \n",
       "\n",
       "                                         opt_learning_rate_warmup_factor  \\\n",
       "submission_label                                                           \n",
       "LBNL-1/corigpu_n64_pt1.6.0                                           1.0   \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                              1.0   \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                              NaN   \n",
       "\n",
       "                                         opt_weight_decay  \n",
       "submission_label                                           \n",
       "LBNL-1/corigpu_n64_pt1.6.0                           0.01  \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed              0.01  \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open               NaN  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "submissions_stats_df_opt = pd.json_normalize(submissions_stats_df['opt_opts'])\n",
    "submissions_stats_df_opt.index = submissions_stats_df.index\n",
    "submissions_stats_df_opt = submissions_stats_df[['global_batch_size', 'opt_name']].join(submissions_stats_df_opt) #.add_prefix('opt.')\n",
    "\n",
    "submissions_stats_df_opt_vis = submissions_stats_df.drop(columns=['submission_short_label', 'runs', 'global_batch_size', 'opt_name', 'opt_opts', 'system']).join(submissions_stats_df_run[['run_stop_epoch.mean', 'run_stop_epoch.std']]).join(submissions_stats_df_opt).set_index('submission_label')\n",
    "visualize_submissions_by_opt(submissions_stats_df_opt_vis, sort_keys=['global_batch_size'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### System throughput performance"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>global_batch_size</th>\n",
       "      <th>opt_name</th>\n",
       "      <th>train_samples</th>\n",
       "      <th>eval_samples</th>\n",
       "      <th>division</th>\n",
       "      <th>submission_long_label</th>\n",
       "      <th>submission_epochs_label</th>\n",
       "      <th>run_epoch_throughput.mean</th>\n",
       "      <th>run_epoch_throughput.std</th>\n",
       "      <th>run_epoch_throughput_per_accelerator.mean</th>\n",
       "      <th>run_epoch_throughput_per_accelerator.std</th>\n",
       "      <th>run_avg_epoch_sample_throughput.mean</th>\n",
       "      <th>run_avg_epoch_sample_throughput.std</th>\n",
       "      <th>run_avg_epoch_sample_throughput_per_accelerator.mean</th>\n",
       "      <th>run_avg_epoch_sample_throughput_per_accelerator.std</th>\n",
       "      <th>run_eval_throughput.mean</th>\n",
       "      <th>run_eval_throughput.std</th>\n",
       "      <th>run_eval_throughput_per_accelerator.mean</th>\n",
       "      <th>run_eval_throughput_per_accelerator.std</th>\n",
       "      <th>run_train_throughput.mean</th>\n",
       "      <th>run_train_throughput.std</th>\n",
       "      <th>run_train_throughput_per_accelerator.mean</th>\n",
       "      <th>run_train_throughput_per_accelerator.std</th>\n",
       "      <th>run_compute_budget</th>\n",
       "      <th>system.submitter</th>\n",
       "      <th>system.division</th>\n",
       "      <th>system.system_type</th>\n",
       "      <th>system.status</th>\n",
       "      <th>system.system_name</th>\n",
       "      <th>system.number_of_nodes</th>\n",
       "      <th>system.host_processors_per_node</th>\n",
       "      <th>system.host_processor_model_name</th>\n",
       "      <th>system.host_processor_core_count</th>\n",
       "      <th>system.host_processor_vcpu_count</th>\n",
       "      <th>system.host_processor_frequency</th>\n",
       "      <th>system.host_processor_caches</th>\n",
       "      <th>system.host_processor_interconnect</th>\n",
       "      <th>system.host_memory_capacity</th>\n",
       "      <th>system.host_storage_type</th>\n",
       "      <th>system.host_storage_capacity</th>\n",
       "      <th>system.host_networking</th>\n",
       "      <th>system.host_networking_topology</th>\n",
       "      <th>system.host_memory_configuration</th>\n",
       "      <th>system.accelerators_per_node</th>\n",
       "      <th>system.accelerator_model_name</th>\n",
       "      <th>system.accelerator_host_interconnect</th>\n",
       "      <th>system.accelerator_frequency</th>\n",
       "      <th>system.accelerator_on-chip_memories</th>\n",
       "      <th>system.accelerator_memory_configuration</th>\n",
       "      <th>system.accelerator_memory_capacity</th>\n",
       "      <th>system.accelerator_interconnect</th>\n",
       "      <th>system.accelerator_interconnect_topology</th>\n",
       "      <th>system.cooling</th>\n",
       "      <th>system.hw_notes</th>\n",
       "      <th>system.framework</th>\n",
       "      <th>system.other_software_stack</th>\n",
       "      <th>system.operating_system</th>\n",
       "      <th>system.sw_notes</th>\n",
       "      <th>system.other_software_stack.CUDA</th>\n",
       "      <th>system.other_software_stack.NCCL</th>\n",
       "      <th>system.other_software_stack.h5py</th>\n",
       "      <th>system.number_of_compute_units</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>submission_label</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>LBNL-1/corigpu_n64_pt1.6.0</th>\n",
       "      <td>128</td>\n",
       "      <td>LAMB</td>\n",
       "      <td>121216</td>\n",
       "      <td>15158</td>\n",
       "      <td>closed</td>\n",
       "      <td>Cori-GPU-64</td>\n",
       "      <td>Cori\\n-GPU-64</td>\n",
       "      <td>0.071793</td>\n",
       "      <td>0.001919</td>\n",
       "      <td>0.001122</td>\n",
       "      <td>0.000030</td>\n",
       "      <td>13636.646533</td>\n",
       "      <td>307.016017</td>\n",
       "      <td>3.551210</td>\n",
       "      <td>0.079952</td>\n",
       "      <td>13605.274039</td>\n",
       "      <td>704.124769</td>\n",
       "      <td>3.543040</td>\n",
       "      <td>0.183366</td>\n",
       "      <td>13655.973840</td>\n",
       "      <td>496.046978</td>\n",
       "      <td>3.556243</td>\n",
       "      <td>0.129179</td>\n",
       "      <td>148.574519</td>\n",
       "      <td>LBNL</td>\n",
       "      <td>closed</td>\n",
       "      <td>NaN</td>\n",
       "      <td>onprem</td>\n",
       "      <td>corigpu_n64_pt1.6.0</td>\n",
       "      <td>8</td>\n",
       "      <td>2</td>\n",
       "      <td>Intel Xeon Gold 6148</td>\n",
       "      <td>20</td>\n",
       "      <td>40</td>\n",
       "      <td>2.4 GHz</td>\n",
       "      <td></td>\n",
       "      <td></td>\n",
       "      <td>384 GB DDR4</td>\n",
       "      <td>on-node NVMe storage</td>\n",
       "      <td>930 GB</td>\n",
       "      <td>4 dual-port Mellanox MT27800 (ConnectX-5) EDR ...</td>\n",
       "      <td></td>\n",
       "      <td></td>\n",
       "      <td>8</td>\n",
       "      <td>NVIDIA V100</td>\n",
       "      <td>PCIe 3.0</td>\n",
       "      <td>1245 MHz</td>\n",
       "      <td></td>\n",
       "      <td></td>\n",
       "      <td>8 GB</td>\n",
       "      <td>NVLink</td>\n",
       "      <td>hybrid cube-mesh</td>\n",
       "      <td></td>\n",
       "      <td>https://docs-dev.nersc.gov/cgpu/hardware/</td>\n",
       "      <td>PyTorch 1.6.0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>openSUSE 15.0</td>\n",
       "      <td></td>\n",
       "      <td>10.2.89</td>\n",
       "      <td>2.5.6</td>\n",
       "      <td>2.10.0</td>\n",
       "      <td>64</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fujitsu-1/abci_1024xV100_pytorch_closed</th>\n",
       "      <td>2048</td>\n",
       "      <td>LAMB</td>\n",
       "      <td>121266</td>\n",
       "      <td>15158</td>\n",
       "      <td>closed</td>\n",
       "      <td>ABCI-1024</td>\n",
       "      <td>ABCI-1024</td>\n",
       "      <td>2.524463</td>\n",
       "      <td>0.007334</td>\n",
       "      <td>0.002465</td>\n",
       "      <td>0.000007</td>\n",
       "      <td>328452.837177</td>\n",
       "      <td>954.164474</td>\n",
       "      <td>5.345912</td>\n",
       "      <td>0.015530</td>\n",
       "      <td>452842.370981</td>\n",
       "      <td>427.222282</td>\n",
       "      <td>7.370481</td>\n",
       "      <td>0.006953</td>\n",
       "      <td>322003.905651</td>\n",
       "      <td>993.958276</td>\n",
       "      <td>5.240949</td>\n",
       "      <td>0.016178</td>\n",
       "      <td>199.777564</td>\n",
       "      <td>Fujitsu</td>\n",
       "      <td>closed</td>\n",
       "      <td>datacenter</td>\n",
       "      <td>available</td>\n",
       "      <td>ABCI PRIMERGY CX2570 M4</td>\n",
       "      <td>256</td>\n",
       "      <td>2</td>\n",
       "      <td>Intel Xeon Gold 6148</td>\n",
       "      <td>40</td>\n",
       "      <td>80</td>\n",
       "      <td>2400MHz</td>\n",
       "      <td>L1: 32KB I + 32KB D per core, L2: 1MB I+D per ...</td>\n",
       "      <td>3x 10.4GT/s UPI</td>\n",
       "      <td>384GB</td>\n",
       "      <td>SSD+NVMe</td>\n",
       "      <td>1600GB</td>\n",
       "      <td>N/A</td>\n",
       "      <td>N/A</td>\n",
       "      <td>12x32GB</td>\n",
       "      <td>4</td>\n",
       "      <td>NVIDIA Tesla V100</td>\n",
       "      <td>PCIe 3.0 x16</td>\n",
       "      <td>1530MHz</td>\n",
       "      <td>L1: 80x 128KB, L2: 6MB per chip</td>\n",
       "      <td>HBM2</td>\n",
       "      <td>16GB</td>\n",
       "      <td>6x 25GT/s NVLink</td>\n",
       "      <td>Direct</td>\n",
       "      <td>Water-cooled</td>\n",
       "      <td>none</td>\n",
       "      <td>PyTorch 1.6.0</td>\n",
       "      <td>cuda 10.2.89, cudnn, 7.6.5, cublas 10.2.2.89, ...</td>\n",
       "      <td>CentOS 7.5.1804</td>\n",
       "      <td>none</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1024</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fujitsu-2/abci_1024xV100_fjpytorch_open</th>\n",
       "      <td>2048</td>\n",
       "      <td>LAMB</td>\n",
       "      <td>121266</td>\n",
       "      <td>15158</td>\n",
       "      <td>open</td>\n",
       "      <td>ABCI-1024</td>\n",
       "      <td>ABCI\\n-1024</td>\n",
       "      <td>2.775585</td>\n",
       "      <td>0.056186</td>\n",
       "      <td>0.002711</td>\n",
       "      <td>0.000055</td>\n",
       "      <td>341240.183868</td>\n",
       "      <td>8540.870875</td>\n",
       "      <td>5.554039</td>\n",
       "      <td>0.139012</td>\n",
       "      <td>281674.493324</td>\n",
       "      <td>53363.772466</td>\n",
       "      <td>4.584546</td>\n",
       "      <td>0.868551</td>\n",
       "      <td>342041.745864</td>\n",
       "      <td>8021.315726</td>\n",
       "      <td>5.567086</td>\n",
       "      <td>0.130555</td>\n",
       "      <td>178.950732</td>\n",
       "      <td>Fujitsu</td>\n",
       "      <td>open</td>\n",
       "      <td>datacenter</td>\n",
       "      <td>available</td>\n",
       "      <td>ABCI PRIMERGY CX2570 M4</td>\n",
       "      <td>256</td>\n",
       "      <td>2</td>\n",
       "      <td>Intel Xeon Gold 6148</td>\n",
       "      <td>40</td>\n",
       "      <td>80</td>\n",
       "      <td>2400MHz</td>\n",
       "      <td>L1: 32KB I + 32KB D per core, L2: 1MB I+D per ...</td>\n",
       "      <td>3x 10.4GT/s UPI</td>\n",
       "      <td>384GB</td>\n",
       "      <td>SSD+NVMe</td>\n",
       "      <td>1600GB</td>\n",
       "      <td>N/A</td>\n",
       "      <td>N/A</td>\n",
       "      <td>12x32GB</td>\n",
       "      <td>4</td>\n",
       "      <td>NVIDIA Tesla V100</td>\n",
       "      <td>PCIe 3.0 x16</td>\n",
       "      <td>1530MHz</td>\n",
       "      <td>L1: 80x 128KB, L2: 6MB per chip</td>\n",
       "      <td>HBM2</td>\n",
       "      <td>16GB</td>\n",
       "      <td>6x 25GT/s NVLink</td>\n",
       "      <td>Direct</td>\n",
       "      <td>Water-cooled</td>\n",
       "      <td>none</td>\n",
       "      <td>PyTorch 1.6.0</td>\n",
       "      <td>cuda 10.2.89, cudnn, 7.6.5, cublas 10.2.2.89, ...</td>\n",
       "      <td>CentOS 7.5.1804</td>\n",
       "      <td>none</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1024</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                         global_batch_size opt_name  ...  system.other_software_stack.h5py  system.number_of_compute_units\n",
       "submission_label                                                     ...                                                                  \n",
       "LBNL-1/corigpu_n64_pt1.6.0                             128     LAMB  ...                            2.10.0                              64\n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed               2048     LAMB  ...                               NaN                            1024\n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open               2048     LAMB  ...                               NaN                            1024\n",
       "\n",
       "[3 rows x 62 columns]"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd.set_option(\"display.max_columns\", 0)\n",
    "submissions_stats_df_system = pd.json_normalize(submissions_stats_df['system']).add_prefix('system.')\n",
    "submissions_stats_df_system['system.number_of_compute_units'] = \\\n",
    "    submissions_stats_df_system.apply(\n",
    "    lambda submission_stat:  int(submission_stat['system.number_of_nodes'])*\n",
    "                             int(submission_stat['system.accelerators_per_node'] \n",
    "                             if int(submission_stat['system.accelerators_per_node']) > 0 \n",
    "                             else submission_stat['system.host_processors_per_node']), axis=1).rename_axis('system.number_of_compute_units')\n",
    "submissions_stats_df_system.index = submissions_stats_df.index\n",
    "\n",
    "throughput_keys = [['run_epoch_throughput.mean','run_epoch_throughput.std'],\n",
    "                   ['run_avg_epoch_sample_throughput.mean', 'run_avg_epoch_sample_throughput.std'],\n",
    "                   ['run_eval_throughput.mean', 'run_eval_throughput.std'],\n",
    "                   ['run_train_throughput.mean', 'run_train_throughput.std']]\n",
    "submissions_stats_df_system_run_vis_list = []\n",
    "\n",
    "for throughput_key in throughput_keys:\n",
    "    submissions_stats_df_system_run_vis = \\\n",
    "        submissions_stats_df_run[throughput_key].join(submissions_stats_df_system)\n",
    "    submissions_stats_df_system_run_vis = \\\n",
    "        submissions_stats_df_run[throughput_key].join(\n",
    "            submissions_stats_df_system_run_vis.apply(lambda submissions_stat: submissions_stat[throughput_key[0]:throughput_key[1]]/\n",
    "                                                                                                    ( float(submissions_stat['system.number_of_nodes'])*\n",
    "                                                                                                      float(submissions_stat['system.accelerators_per_node'] \n",
    "                                                                                                            if int(submissions_stat['system.accelerators_per_node']) > 0 \n",
    "                                                                                                            else submissions_stat['system.host_processors_per_node']) )/(1. if throughput_key[0].startswith('run_epoch') else 60.), axis=1).rename(columns=lambda s: '_per_accelerator.'.join(s.split('.'))))\n",
    "    submissions_stats_df_system_run_vis_list.append(submissions_stats_df_system_run_vis)\n",
    "submissions_stats_df_system_run_vis = pd.concat(submissions_stats_df_system_run_vis_list, axis=1)\n",
    "submissions_stats_df_system_run_vis['run_compute_budget'] = submissions_stats_df_system['system.number_of_compute_units']*submissions_stats_df_run['run_time.mean'].apply(lambda t: t/60.)\n",
    "\n",
    "submissions_stats_df_system_vis = submissions_stats_df.drop(columns=['submission_short_label', 'runs', 'opt_opts', 'system']).join(submissions_stats_df_system_run_vis).join(submissions_stats_df_system).set_index('submission_label')\n",
    "# visualize_submissions_by_opt(submissions_stats_df_system_vis, sort_keys=['run_epoch_throughput_per_accelerator.mean'])\n",
    "submissions_stats_df_system_vis.sort_values(['division', 'global_batch_size']) # 'run_epoch_throughput_per_accelerator.mean'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Systems table"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>system.number_of_nodes</th>\n",
       "      <th>system.host_processors_per_node</th>\n",
       "      <th>system.host_processor_model_name</th>\n",
       "      <th>system.accelerators_per_node</th>\n",
       "      <th>system.accelerator_model_name</th>\n",
       "      <th>system.accelerator_memory_capacity</th>\n",
       "      <th>system.host_memory_capacity</th>\n",
       "      <th>system.host_storage_type</th>\n",
       "      <th>system.host_storage_capacity</th>\n",
       "      <th>system.host_networking</th>\n",
       "      <th>system.host_networking_topology</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>submission_label</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Fujitsu-1/abci_1024xV100_pytorch_closed</th>\n",
       "      <td>256</td>\n",
       "      <td>2</td>\n",
       "      <td>Intel Xeon Gold 6148</td>\n",
       "      <td>4</td>\n",
       "      <td>NVIDIA Tesla V100</td>\n",
       "      <td>16GB</td>\n",
       "      <td>384GB</td>\n",
       "      <td>SSD+NVMe</td>\n",
       "      <td>1600GB</td>\n",
       "      <td>N/A</td>\n",
       "      <td>N/A</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LBNL-1/corigpu_n64_pt1.6.0</th>\n",
       "      <td>8</td>\n",
       "      <td>2</td>\n",
       "      <td>Intel Xeon Gold 6148</td>\n",
       "      <td>8</td>\n",
       "      <td>NVIDIA V100</td>\n",
       "      <td>8 GB</td>\n",
       "      <td>384 GB DDR4</td>\n",
       "      <td>on-node NVMe storage</td>\n",
       "      <td>930 GB</td>\n",
       "      <td>4 dual-port Mellanox MT27800 (ConnectX-5) EDR ...</td>\n",
       "      <td></td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fujitsu-2/abci_1024xV100_fjpytorch_open</th>\n",
       "      <td>256</td>\n",
       "      <td>2</td>\n",
       "      <td>Intel Xeon Gold 6148</td>\n",
       "      <td>4</td>\n",
       "      <td>NVIDIA Tesla V100</td>\n",
       "      <td>16GB</td>\n",
       "      <td>384GB</td>\n",
       "      <td>SSD+NVMe</td>\n",
       "      <td>1600GB</td>\n",
       "      <td>N/A</td>\n",
       "      <td>N/A</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                        system.number_of_nodes  ... system.host_networking_topology\n",
       "submission_label                                                ...                                \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed                    256  ...                             N/A\n",
       "LBNL-1/corigpu_n64_pt1.6.0                                   8  ...                                \n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open                    256  ...                             N/A\n",
       "\n",
       "[3 rows x 11 columns]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# System configuration\n",
    "display(submissions_stats_df_system_vis[[ 'system.number_of_nodes', 'system.host_processors_per_node', 'system.host_processor_model_name', 'system.accelerators_per_node', 'system.accelerator_model_name', 'system.accelerator_memory_capacity', 'system.host_memory_capacity',  'system.host_storage_type', 'system.host_storage_capacity', 'system.host_networking', 'system.host_networking_topology']])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Compute budget (hours x # compute units)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>run_compute_budget</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>submission_label</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Fujitsu-1/abci_1024xV100_pytorch_closed</th>\n",
       "      <td>199.777564</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LBNL-1/corigpu_n64_pt1.6.0</th>\n",
       "      <td>148.574519</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fujitsu-2/abci_1024xV100_fjpytorch_open</th>\n",
       "      <td>178.950732</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                         run_compute_budget\n",
       "submission_label                                           \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed          199.777564\n",
       "LBNL-1/corigpu_n64_pt1.6.0                       148.574519\n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open          178.950732"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Compute budget\n",
    "submissions_stats_df_system_vis[['run_compute_budget']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cross-benchmark compute budget comparison\n",
    "if benchmark.round == 'hpc_results_v0.7' and benchmark.name == 'cosmoflow':\n",
    "    submissions_stats_df_system_vis.loc[['LBNL-3/corigpu_n64_tf1.15.0','Fujitsu-4/abci_512xV100_tensorflow_closed','Fujitsu-1/abci_2048xV100_tensorflow_open'],'run_compute_budget']/np.array([148.575, 199.778, 178.951])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Explaining run-to-run variation in terms of # epochs vs. system throughput with logarithmic PCA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>explained_variance_1</th>\n",
       "      <th>explained_variance_ratio_1</th>\n",
       "      <th>epochs_component_1</th>\n",
       "      <th>system_component_1</th>\n",
       "      <th>explained_variance_2</th>\n",
       "      <th>explained_variance_ratio_2</th>\n",
       "      <th>epochs_component_2</th>\n",
       "      <th>system_component_2</th>\n",
       "      <th>epochs_system_correlation</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>submission_label</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Fujitsu-1/abci_1024xV100_pytorch_closed</th>\n",
       "      <td>0.000002</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.558375e-26</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>7.395571e-32</td>\n",
       "      <td>4.676760e-26</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>-1.558375e-26</td>\n",
       "      <td>7.208606e-14</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fujitsu-2/abci_1024xV100_fjpytorch_open</th>\n",
       "      <td>0.000495</td>\n",
       "      <td>0.967857</td>\n",
       "      <td>9.376351e-01</td>\n",
       "      <td>-0.347621</td>\n",
       "      <td>1.643679e-05</td>\n",
       "      <td>3.214297e-02</td>\n",
       "      <td>-0.347621</td>\n",
       "      <td>-9.376351e-01</td>\n",
       "      <td>-8.656630e-01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LBNL-1/corigpu_n64_pt1.6.0</th>\n",
       "      <td>0.000128</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                         explained_variance_1  explained_variance_ratio_1  ...  system_component_2  epochs_system_correlation\n",
       "submission_label                                                                           ...                                               \n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed              0.000002                    1.000000  ...       -1.558375e-26               7.208606e-14\n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open              0.000495                    0.967857  ...       -9.376351e-01              -8.656630e-01\n",
       "LBNL-1/corigpu_n64_pt1.6.0                           0.000128                    1.000000  ...        0.000000e+00                        NaN\n",
       "\n",
       "[3 rows x 9 columns]"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.decomposition import PCA\n",
    "\n",
    "log_pca = dict({k:[] for k in ['submission_label',\n",
    "                               'explained_variance_1', 'explained_variance_ratio_1','epochs_component_1','system_component_1',\n",
    "                               'explained_variance_2', 'explained_variance_ratio_2','epochs_component_2','system_component_2',\n",
    "                               'epochs_system_correlation']})\n",
    "for k, submission_stat in submissions_stats.items():\n",
    "    log_pca['submission_label'].append(submission_stat['submission_label'])\n",
    "    pca = PCA(n_components=2)\n",
    "    log_num_epochs_log_epoch_times = pd.DataFrame(sorted(submissions_stats[k]['runs'], key=lambda run: run['run_time'])[1:mlperf_opts[benchmark.round][benchmark.name]['runs_to_submit']-1])[['run_stop_epoch', 'run_epoch_time']].applymap(np.log10)\n",
    "    pca.fit(log_num_epochs_log_epoch_times)\n",
    "    for i in range(2):\n",
    "        log_pca[f'explained_variance_{i+1}'].append(pca.explained_variance_[i])\n",
    "        log_pca[f'explained_variance_ratio_{i+1}'].append(pca.explained_variance_ratio_[i])\n",
    "        log_pca[f'epochs_component_{i+1}'].append(pca.components_[i][0])\n",
    "        log_pca[f'system_component_{i+1}'].append(pca.components_[i][1])\n",
    "    log_pca[f'epochs_system_correlation'].append(log_num_epochs_log_epoch_times.corr().iloc[0,1])\n",
    "run_to_run_variation_log_pca_df = pd.DataFrame.from_dict(log_pca).set_index('submission_label')\n",
    "run_to_run_variation_log_pca_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "def generate_run_to_run_variation_log_pca_plot():\n",
    "    # Log-PCA plot\n",
    "    fig, ax  = plt.subplots(nrows=1, ncols=1, figsize=(19.5, 6.5))\n",
    "    colors = create_submission_colors(color_label=accelerator_short_label, color_label_range=accelerator_short_names)    \n",
    "    run_to_run_variation_log_pca_acc_df = \\\n",
    "        pd.concat([run_to_run_variation_log_pca_df, \n",
    "                   submissions_stats_df_system_vis[['system.host_processors_per_node', 'system.host_processor_model_name', 'system.accelerators_per_node', 'system.accelerator_model_name']].apply(\n",
    "                       lambda r: accelerator_short_label(r['system.accelerator_model_name'] if int(r['system.accelerators_per_node']) > 0 else r['system.host_processor_model_name']), axis=1).rename('accelerator_short_label')], axis=1)\n",
    "\n",
    "    ax_box_xlim =(np.inf, -np.inf)\n",
    "    ax_box_ylim =(np.inf, -np.inf)\n",
    "\n",
    "    for s_label, s_log_pca_acc in run_to_run_variation_log_pca_acc_df.iterrows():\n",
    "\n",
    "        half_axis_log = []\n",
    "        for i in range(1,3):\n",
    "            half_axis_log.append(np.array(\n",
    "                     [np.sqrt(s_log_pca_acc[f'explained_variance_{i}'])*s_log_pca_acc[f'epochs_component_{i}']*np.sign(s_log_pca_acc[f'epochs_component_{i}']),\n",
    "                      np.sqrt(s_log_pca_acc[f'explained_variance_{i}'])*s_log_pca_acc[f'system_component_{i}']*np.sign(s_log_pca_acc[f'epochs_component_{i}'])]))\n",
    "\n",
    "        t = np.linspace(0,2.*np.pi, 1000)\n",
    "        circle_coord = [10**(half_axis_log[0][0]*np.cos(t)+half_axis_log[1][0]*np.sin(t)), \n",
    "                        10**(half_axis_log[0][1]*np.cos(t)+half_axis_log[1][1]*np.sin(t))]\n",
    "        ax.loglog(circle_coord[0], circle_coord[1],  color=colors[s_log_pca_acc['accelerator_short_label']], alpha=0.5)\n",
    "        ax_box_xlim = (min(np.min(circle_coord[0]), ax_box_xlim[0]), max(np.max(circle_coord[0]), ax_box_ylim[0]))\n",
    "        ax_box_ylim = (min(np.min(circle_coord[1]), ax_box_ylim[1]), max(np.max(circle_coord[1]), ax_box_ylim[1]))\n",
    "\n",
    "        annotation_pos = np.array([circle_coord[0][0],circle_coord[1][0]])\n",
    "        submission_long_label = submissions_stats_df.set_index('submission_label').loc[s_label]['submission_long_label']\n",
    "        for i in range(2):\n",
    "            ax.loglog(10**(half_axis_log[i][0]*(np.array([-1., 0.] if 'Fugaku' in submission_long_label or 'Cori-GPU' in submission_long_label else np.array([1., 0.]) ) )),\n",
    "                      10**(half_axis_log[i][1]*(np.array([-1., 0.] if 'Fugaku' in submission_long_label or 'Cori-GPU' in submission_long_label else np.array([1., 0.]) ) )),\n",
    "                      '--' if i == 0 else ':', alpha=0.5, color=colors[s_log_pca_acc['accelerator_short_label']])\n",
    "            scatteropts = dict()\n",
    "            if i == 1:\n",
    "                scatteropts['facecolors'] = 'white'\n",
    "            ax.scatter(10**(half_axis_log[i][0]*(-1. if 'Fugaku' in submission_long_label or 'Cori-GPU' in submission_long_label else 1.)),\n",
    "                       10**(half_axis_log[i][1]*(-1. if 'Fugaku' in submission_long_label or 'Cori-GPU' in submission_long_label else 1.)),\n",
    "                       marker='o' if i == 0 else 'D', alpha=.6, **scatteropts, color=colors[s_log_pca_acc['accelerator_short_label']])\n",
    "                    \n",
    "        if 'Fugaku' in submission_long_label or 'Cori-GPU' in submission_long_label:\n",
    "            annotation_circle = annotation_pos \n",
    "            annotation_pos = np.array([annotation_pos[0]**1.1, annotation_pos[1]**2.5])\n",
    "            if 'Fugaku-16384' in submission_long_label:\n",
    "                annotation_pos[1] = annotation_pos[1]**12\n",
    "            elif 'Fugaku-8192' in submission_long_label:\n",
    "                annotation_pos[1] = annotation_pos[1]**12\n",
    "            elif 'Fugaku-512' in submission_long_label:\n",
    "                annotation_pos[1] = annotation_pos[1]**14\n",
    "            elif 'Cori-GPU' in submission_long_label and benchmark.name == 'cosmoflow':\n",
    "#                 annotation_pos = np.array([annotation_pos[0]**1.1, annotation_pos[1]**5])\n",
    "                annotation_pos[1] = annotation_pos[1]**5\n",
    "                \n",
    "            annotation_circle = 1./annotation_circle\n",
    "            annotation_pos = 1./annotation_pos\n",
    "            xytext = (-2, 0)\n",
    "        else:\n",
    "            annotation_circle = annotation_pos*1.\n",
    "            annotation_pos = np.array([annotation_pos[0]**1.1, annotation_pos[1]**2.5 if benchmark.name == 'cosmoflow' else annotation_pos[1]**1.1])\n",
    "            if 'ABCI-2048' in submission_long_label and benchmark.name == 'cosmoflow':\n",
    "                annotation_pos[1] = annotation_pos[1]**-5.5\n",
    "            if 'HAL-64' in submission_long_label:\n",
    "                annotation_pos = np.array([annotation_pos[0]**1.3, annotation_pos[1]**4])\n",
    "            xytext = (2, 0)\n",
    "            \n",
    "        submission_label_annotation_opts = dict(\n",
    "            xytext=xytext,\n",
    "            textcoords=\"offset points\", \n",
    "            ha='left' if ('Fugaku' not in submission_long_label and 'Cori-GPU' not in submission_long_label) else 'right', va='center', \n",
    "            rotation=0,\n",
    "            alpha=1., fontsize=13)\n",
    "        annotation_text = (r'$\\bigstar$' if submissions_stats_df.set_index('submission_label').loc[s_label]['division'] == 'open' else '') + \\\n",
    "                          submission_long_label\n",
    "\n",
    "        ax.plot([annotation_circle[0], annotation_pos[0]], \n",
    "                [annotation_circle[1], annotation_pos[1]], \n",
    "                '-',\n",
    "                color=colors[s_log_pca_acc['accelerator_short_label']],\n",
    "                alpha=0.3\n",
    "               )\n",
    "        ax.annotate(annotation_text,\n",
    "            xy=annotation_pos, # if ('Fugaku' not in submission_long_label and 'Cori-GPU' not in submission_long_label) else 1./annotation_pos, \n",
    "            color=colors[s_log_pca_acc['accelerator_short_label']],\n",
    "            **submission_label_annotation_opts)\n",
    "\n",
    "    ax.annotate('compute time to solution\\n' + r'($\\nwarrow$ low, $\\searrow$ high, ' + '\\n' + r' const. along $\\nearrow$)',\n",
    "                xy=(ax_box_xlim[0], ax_box_ylim[1]),\n",
    "                xytext=(15,-50),\n",
    "                textcoords=\"offset points\",\n",
    "                ha='center', va='bottom',\n",
    "                fontsize='small',\n",
    "                color='black',\n",
    "                alpha=0.5)        \n",
    "\n",
    "#     # Symbolic legend\n",
    "#     symbolic_legend_labels = []\n",
    "#     ax_twin = ax.twinx()\n",
    "\n",
    "#     ax_twin.scatter([], [], label=\"Symbolic legend\", marker=\"None\") \n",
    "#     symbolic_legend_labels.append(\"Symbolic legend\")\n",
    "#     ax_twin.plot([], [], 'o', label='1. principal component', color='black')\n",
    "#     symbolic_legend_labels.append('1. principal component')\n",
    "#     ax_twin.plot([], [], 'D', fillstyle='none', label='2. principal component', color='black')\n",
    "#     symbolic_legend_labels.append('2. principal component')\n",
    "\n",
    "#     ax_twin.get_yaxis().set_visible(False)\n",
    "\n",
    "#     legend_side='right'\n",
    "    \n",
    "#     legend_handles, legend_labels = ax_twin.get_legend_handles_labels()\n",
    "#     legend_labels, legend_handles = zip(*sorted(zip(legend_labels, legend_handles), key=lambda t: symbolic_legend_labels.index(t[0])))\n",
    "#     ax_twin.legend(legend_handles, legend_labels, bbox_to_anchor=( 1.5 if legend_side == 'right' else -0.15, 0), loc='lower right', frameon=False, fontsize='small')\n",
    "    \n",
    "    ax.set_xlabel('# epochs to solution (relative to mean)')\n",
    "    ax.set_ylabel('epoch throughput (relative to mean)')\n",
    "    ax.grid(True, which='both', alpha=0.3)\n",
    "    ax.xaxis.set_major_formatter(plt.ScalarFormatter())\n",
    "    ax.yaxis.set_major_formatter(plt.ScalarFormatter())\n",
    "    ax.xaxis.set_minor_formatter(plt.ScalarFormatter())\n",
    "    ax.yaxis.set_minor_formatter(plt.ScalarFormatter())\n",
    "    ax.set_xlim(10**(np.log10(ax.get_xlim())*1.2))\n",
    "    ax.set_ylim(10**(np.log10(ax.get_ylim())*1.2))\n",
    "    ax.set_aspect('equal')\n",
    "    \n",
    "    fig.tight_layout()\n",
    "# generate_run_to_run_variation_log_pca_plot()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### The complete statistics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>train_samples</th>\n",
       "      <th>eval_samples</th>\n",
       "      <th>division</th>\n",
       "      <th>submission_long_label</th>\n",
       "      <th>submission_epochs_label</th>\n",
       "      <th>eval_time/run_time</th>\n",
       "      <th>run_avg_epoch_sample_throughput.mean</th>\n",
       "      <th>run_avg_epoch_sample_throughput.std</th>\n",
       "      <th>run_avg_epoch_samples.mean</th>\n",
       "      <th>run_avg_epoch_samples.std</th>\n",
       "      <th>run_epoch_throughput.mean</th>\n",
       "      <th>run_epoch_throughput.std</th>\n",
       "      <th>run_epoch_time.mean</th>\n",
       "      <th>run_epoch_time.std</th>\n",
       "      <th>run_eval_throughput.mean</th>\n",
       "      <th>run_eval_throughput.std</th>\n",
       "      <th>run_eval_time.mean</th>\n",
       "      <th>run_eval_time.std</th>\n",
       "      <th>run_num_evals.mean</th>\n",
       "      <th>run_num_evals.std</th>\n",
       "      <th>run_staging_throughput.mean</th>\n",
       "      <th>run_staging_throughput.std</th>\n",
       "      <th>run_staging_time.mean</th>\n",
       "      <th>run_staging_time.std</th>\n",
       "      <th>run_stop_epoch.mean</th>\n",
       "      <th>run_stop_epoch.std</th>\n",
       "      <th>run_time.mean</th>\n",
       "      <th>run_time.std</th>\n",
       "      <th>run_train_throughput.mean</th>\n",
       "      <th>run_train_throughput.std</th>\n",
       "      <th>run_train_time.mean</th>\n",
       "      <th>run_train_time.std</th>\n",
       "      <th>staging_time/run_time</th>\n",
       "      <th>train_time/run_time</th>\n",
       "      <th>global_batch_size</th>\n",
       "      <th>opt_name</th>\n",
       "      <th>opt_epsilon</th>\n",
       "      <th>opt_base_learning_rate</th>\n",
       "      <th>opt_learning_rate_warmup_steps</th>\n",
       "      <th>opt_learning_rate_warmup_factor</th>\n",
       "      <th>opt_weight_decay</th>\n",
       "      <th>system.submitter</th>\n",
       "      <th>system.division</th>\n",
       "      <th>system.system_type</th>\n",
       "      <th>system.status</th>\n",
       "      <th>system.system_name</th>\n",
       "      <th>system.number_of_nodes</th>\n",
       "      <th>system.host_processors_per_node</th>\n",
       "      <th>system.host_processor_model_name</th>\n",
       "      <th>system.host_processor_core_count</th>\n",
       "      <th>system.host_processor_vcpu_count</th>\n",
       "      <th>system.host_processor_frequency</th>\n",
       "      <th>system.host_processor_caches</th>\n",
       "      <th>system.host_processor_interconnect</th>\n",
       "      <th>system.host_memory_capacity</th>\n",
       "      <th>system.host_storage_type</th>\n",
       "      <th>system.host_storage_capacity</th>\n",
       "      <th>system.host_networking</th>\n",
       "      <th>system.host_networking_topology</th>\n",
       "      <th>system.host_memory_configuration</th>\n",
       "      <th>system.accelerators_per_node</th>\n",
       "      <th>system.accelerator_model_name</th>\n",
       "      <th>system.accelerator_host_interconnect</th>\n",
       "      <th>system.accelerator_frequency</th>\n",
       "      <th>system.accelerator_on-chip_memories</th>\n",
       "      <th>system.accelerator_memory_configuration</th>\n",
       "      <th>system.accelerator_memory_capacity</th>\n",
       "      <th>system.accelerator_interconnect</th>\n",
       "      <th>system.accelerator_interconnect_topology</th>\n",
       "      <th>system.cooling</th>\n",
       "      <th>system.hw_notes</th>\n",
       "      <th>system.framework</th>\n",
       "      <th>system.other_software_stack</th>\n",
       "      <th>system.operating_system</th>\n",
       "      <th>system.sw_notes</th>\n",
       "      <th>system.other_software_stack.CUDA</th>\n",
       "      <th>system.other_software_stack.NCCL</th>\n",
       "      <th>system.other_software_stack.h5py</th>\n",
       "      <th>system.number_of_compute_units</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>submission_label</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>LBNL-1/corigpu_n64_pt1.6.0</th>\n",
       "      <td>121216</td>\n",
       "      <td>15158</td>\n",
       "      <td>closed</td>\n",
       "      <td>Cori-GPU-64</td>\n",
       "      <td>Cori\\n-GPU-64</td>\n",
       "      <td>0.362732</td>\n",
       "      <td>13636.646533</td>\n",
       "      <td>307.016017</td>\n",
       "      <td>189932.0</td>\n",
       "      <td>874.685658</td>\n",
       "      <td>0.071793</td>\n",
       "      <td>0.001919</td>\n",
       "      <td>13.928849</td>\n",
       "      <td>0.362650</td>\n",
       "      <td>13605.274039</td>\n",
       "      <td>704.124769</td>\n",
       "      <td>50.524406</td>\n",
       "      <td>3.044168</td>\n",
       "      <td>45.333333</td>\n",
       "      <td>0.577350</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>10.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>139.288611</td>\n",
       "      <td>3.626479</td>\n",
       "      <td>13655.973840</td>\n",
       "      <td>496.046978</td>\n",
       "      <td>88.764083</td>\n",
       "      <td>3.111298</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.637267</td>\n",
       "      <td>128</td>\n",
       "      <td>LAMB</td>\n",
       "      <td>1.000000e-08</td>\n",
       "      <td>0.0010</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.01</td>\n",
       "      <td>LBNL</td>\n",
       "      <td>closed</td>\n",
       "      <td>NaN</td>\n",
       "      <td>onprem</td>\n",
       "      <td>corigpu_n64_pt1.6.0</td>\n",
       "      <td>8</td>\n",
       "      <td>2</td>\n",
       "      <td>Intel Xeon Gold 6148</td>\n",
       "      <td>20</td>\n",
       "      <td>40</td>\n",
       "      <td>2.4 GHz</td>\n",
       "      <td></td>\n",
       "      <td></td>\n",
       "      <td>384 GB DDR4</td>\n",
       "      <td>on-node NVMe storage</td>\n",
       "      <td>930 GB</td>\n",
       "      <td>4 dual-port Mellanox MT27800 (ConnectX-5) EDR ...</td>\n",
       "      <td></td>\n",
       "      <td></td>\n",
       "      <td>8</td>\n",
       "      <td>NVIDIA V100</td>\n",
       "      <td>PCIe 3.0</td>\n",
       "      <td>1245 MHz</td>\n",
       "      <td></td>\n",
       "      <td></td>\n",
       "      <td>8 GB</td>\n",
       "      <td>NVLink</td>\n",
       "      <td>hybrid cube-mesh</td>\n",
       "      <td></td>\n",
       "      <td>https://docs-dev.nersc.gov/cgpu/hardware/</td>\n",
       "      <td>PyTorch 1.6.0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>openSUSE 15.0</td>\n",
       "      <td></td>\n",
       "      <td>10.2.89</td>\n",
       "      <td>2.5.6</td>\n",
       "      <td>2.10.0</td>\n",
       "      <td>64</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fujitsu-1/abci_1024xV100_pytorch_closed</th>\n",
       "      <td>121266</td>\n",
       "      <td>15158</td>\n",
       "      <td>closed</td>\n",
       "      <td>ABCI-1024</td>\n",
       "      <td>ABCI-1024</td>\n",
       "      <td>0.040034</td>\n",
       "      <td>328452.837177</td>\n",
       "      <td>954.164474</td>\n",
       "      <td>130108.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>2.524463</td>\n",
       "      <td>0.007334</td>\n",
       "      <td>0.396124</td>\n",
       "      <td>0.001147</td>\n",
       "      <td>452842.370981</td>\n",
       "      <td>427.222282</td>\n",
       "      <td>0.468622</td>\n",
       "      <td>0.000442</td>\n",
       "      <td>14.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>62072.354483</td>\n",
       "      <td>180.453773</td>\n",
       "      <td>2.197822</td>\n",
       "      <td>0.006371</td>\n",
       "      <td>24.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>11.705717</td>\n",
       "      <td>0.021555</td>\n",
       "      <td>322003.905651</td>\n",
       "      <td>993.958276</td>\n",
       "      <td>9.038350</td>\n",
       "      <td>0.027814</td>\n",
       "      <td>0.187756</td>\n",
       "      <td>0.772131</td>\n",
       "      <td>2048</td>\n",
       "      <td>LAMB</td>\n",
       "      <td>1.000000e-08</td>\n",
       "      <td>0.0055</td>\n",
       "      <td>400.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.01</td>\n",
       "      <td>Fujitsu</td>\n",
       "      <td>closed</td>\n",
       "      <td>datacenter</td>\n",
       "      <td>available</td>\n",
       "      <td>ABCI PRIMERGY CX2570 M4</td>\n",
       "      <td>256</td>\n",
       "      <td>2</td>\n",
       "      <td>Intel Xeon Gold 6148</td>\n",
       "      <td>40</td>\n",
       "      <td>80</td>\n",
       "      <td>2400MHz</td>\n",
       "      <td>L1: 32KB I + 32KB D per core, L2: 1MB I+D per ...</td>\n",
       "      <td>3x 10.4GT/s UPI</td>\n",
       "      <td>384GB</td>\n",
       "      <td>SSD+NVMe</td>\n",
       "      <td>1600GB</td>\n",
       "      <td>N/A</td>\n",
       "      <td>N/A</td>\n",
       "      <td>12x32GB</td>\n",
       "      <td>4</td>\n",
       "      <td>NVIDIA Tesla V100</td>\n",
       "      <td>PCIe 3.0 x16</td>\n",
       "      <td>1530MHz</td>\n",
       "      <td>L1: 80x 128KB, L2: 6MB per chip</td>\n",
       "      <td>HBM2</td>\n",
       "      <td>16GB</td>\n",
       "      <td>6x 25GT/s NVLink</td>\n",
       "      <td>Direct</td>\n",
       "      <td>Water-cooled</td>\n",
       "      <td>none</td>\n",
       "      <td>PyTorch 1.6.0</td>\n",
       "      <td>cuda 10.2.89, cudnn, 7.6.5, cublas 10.2.2.89, ...</td>\n",
       "      <td>CentOS 7.5.1804</td>\n",
       "      <td>none</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1024</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fujitsu-2/abci_1024xV100_fjpytorch_open</th>\n",
       "      <td>121266</td>\n",
       "      <td>15158</td>\n",
       "      <td>open</td>\n",
       "      <td>ABCI-1024</td>\n",
       "      <td>ABCI\\n-1024</td>\n",
       "      <td>0.013057</td>\n",
       "      <td>341240.183868</td>\n",
       "      <td>8540.870875</td>\n",
       "      <td>122953.0</td>\n",
       "      <td>639.126748</td>\n",
       "      <td>2.775585</td>\n",
       "      <td>0.056186</td>\n",
       "      <td>0.360284</td>\n",
       "      <td>0.007149</td>\n",
       "      <td>281674.493324</td>\n",
       "      <td>53363.772466</td>\n",
       "      <td>0.136906</td>\n",
       "      <td>0.033568</td>\n",
       "      <td>2.666667</td>\n",
       "      <td>1.154701</td>\n",
       "      <td>69521.318153</td>\n",
       "      <td>2846.825395</td>\n",
       "      <td>1.962333</td>\n",
       "      <td>0.077194</td>\n",
       "      <td>23.666667</td>\n",
       "      <td>1.154701</td>\n",
       "      <td>10.485394</td>\n",
       "      <td>0.228172</td>\n",
       "      <td>342041.745864</td>\n",
       "      <td>8021.315726</td>\n",
       "      <td>8.385078</td>\n",
       "      <td>0.246814</td>\n",
       "      <td>0.187149</td>\n",
       "      <td>0.799691</td>\n",
       "      <td>2048</td>\n",
       "      <td>LAMB</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Fujitsu</td>\n",
       "      <td>open</td>\n",
       "      <td>datacenter</td>\n",
       "      <td>available</td>\n",
       "      <td>ABCI PRIMERGY CX2570 M4</td>\n",
       "      <td>256</td>\n",
       "      <td>2</td>\n",
       "      <td>Intel Xeon Gold 6148</td>\n",
       "      <td>40</td>\n",
       "      <td>80</td>\n",
       "      <td>2400MHz</td>\n",
       "      <td>L1: 32KB I + 32KB D per core, L2: 1MB I+D per ...</td>\n",
       "      <td>3x 10.4GT/s UPI</td>\n",
       "      <td>384GB</td>\n",
       "      <td>SSD+NVMe</td>\n",
       "      <td>1600GB</td>\n",
       "      <td>N/A</td>\n",
       "      <td>N/A</td>\n",
       "      <td>12x32GB</td>\n",
       "      <td>4</td>\n",
       "      <td>NVIDIA Tesla V100</td>\n",
       "      <td>PCIe 3.0 x16</td>\n",
       "      <td>1530MHz</td>\n",
       "      <td>L1: 80x 128KB, L2: 6MB per chip</td>\n",
       "      <td>HBM2</td>\n",
       "      <td>16GB</td>\n",
       "      <td>6x 25GT/s NVLink</td>\n",
       "      <td>Direct</td>\n",
       "      <td>Water-cooled</td>\n",
       "      <td>none</td>\n",
       "      <td>PyTorch 1.6.0</td>\n",
       "      <td>cuda 10.2.89, cudnn, 7.6.5, cublas 10.2.2.89, ...</td>\n",
       "      <td>CentOS 7.5.1804</td>\n",
       "      <td>none</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1024</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                         train_samples  eval_samples  ... system.other_software_stack.h5py system.number_of_compute_units\n",
       "submission_label                                                      ...                                                                \n",
       "LBNL-1/corigpu_n64_pt1.6.0                      121216         15158  ...                           2.10.0                             64\n",
       "Fujitsu-1/abci_1024xV100_pytorch_closed         121266         15158  ...                              NaN                           1024\n",
       "Fujitsu-2/abci_1024xV100_fjpytorch_open         121266         15158  ...                              NaN                           1024\n",
       "\n",
       "[3 rows x 79 columns]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "submissions_stats_df_vis = submissions_stats_df.drop(columns=['submission_short_label', 'runs', 'global_batch_size', 'opt_name', 'opt_opts', 'system']).join(submissions_stats_df_run).join(submissions_stats_df_opt).join(submissions_stats_df_system).set_index('submission_label').sort_values(['division', 'global_batch_size'])\n",
    "visualize_submissions_by_opt(submissions_stats_df_vis)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Visual analysis"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Time to solution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1296x864 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "generate_time_to_solution_plot()\n",
    "# plt.savefig(f'output/{benchmark.name}-time-to-solution.png')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Relative time analysis: data staging, evaluation and training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x81 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "generate_relative_staging_eval_training_times_plot()\n",
    "# plt.savefig('output/results-relative-staging-eval-training-times.png', dpi=300)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Compute time analysis: Parameter choice (batch size, # compute units), epoch- and throughput scaling, compute time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Note: different number of training samples used in different submissions! Using median of [121216 121266]\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 2304x1224 with 5 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "generate_compute_time_analysis_plot()\n",
    "# plt.savefig(f'output/{benchmark.name}-2x2-compute-analysis-batch-size-acc-epochs-throughput-efficiency-vertical.png', dpi=300)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Explaining run-to-run variance in terms of # epochs vs. system throughput (logarithmic PCA)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1404x468 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# half-axis correspond to sqrt(variance) of principal components (dashed line first component, dotted line second component)\n",
    "generate_run_to_run_variation_log_pca_plot()\n",
    "# plt.savefig(f'output/{benchmark.name}-log-pca-epochs-throughput.png', dpi=300)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}