notebooks/model-test.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "8f937365-15c3-41b4-89c5-62ad62771999",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/home/tk/repos/erc\n"
     ]
    }
   ],
   "source": [
    "%cd /home/tk/repos/erc/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "f0941f3c-16c9-41a6-9f78-cba4bb356913",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "('IEMOCAP', 'upper', 1000, 0, 4)"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ROOT_DIR = \"./multimodal-datasets/\"\n",
    "# model_checkpoint= '/home/tk/repos/erc/results/MELD/roberta-large/SEEDS/2021-05-08-19-57-31-speaker_mode-upper-num_past_utterances-1000-num_future_utterances-1000-batch_size-4-seed-4/checkpoint-9992/'\n",
    "# model_checkpoint = '/home/tk/repos/erc/results/IEMOCAP/roberta-large/SEEDS/2021-05-08-16-15-19-speaker_mode-upper-num_past_utterances-1000-num_future_utterances-1000-batch_size-4-seed-2/checkpoint-5975/'\n",
    "model_checkpoint = \"/home/tk/repos/erc/results/IEMOCAP/roberta-large/SEEDS/2021-05-09-12-19-54-speaker_mode-upper-num_past_utterances-1000-num_future_utterances-0-batch_size-4-seed-4/checkpoint-5975/\"\n",
    "\n",
    "DATASET = model_checkpoint.split(\"/\")[6]\n",
    "params = model_checkpoint.split(\"/\")[9]\n",
    "speaker_mode = params.split(\"-\")[7]\n",
    "num_past_utterances = int(params.split(\"-\")[9])\n",
    "num_future_utterances = int(params.split(\"-\")[11])\n",
    "SEED = int(params.split(\"-\")[-1])\n",
    "\n",
    "DATASET, speaker_mode, num_past_utterances, num_future_utterances, SEED"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "900e9d77-d234-4daf-8489-37bbede007a7",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2021-05-17 23:34:13.892 INFO utils - _load_utterance_ordered: every text file exists fine!\n",
      "2021-05-17 23:34:13.893 INFO utils - _string2tokens: converting utterances into tokens ...\n",
      "2021-05-17 23:34:13.894 INFO utils - _string2tokens: creating input utterance data ... \n",
      "100%|██████████| 31/31 [00:01<00:00, 21.85it/s]\n",
      "2021-05-17 23:34:17.701 INFO utils - _create_input: number of truncated utterances: 1131\n"
     ]
    }
   ],
   "source": [
    "from transformers import AutoTokenizer, AutoModelForSequenceClassification\n",
    "from utils import get_num_classes, ErcTextDataset\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "NUM_CLASSES = get_num_classes(DATASET)\n",
    "tokenizer = AutoTokenizer.from_pretrained(model_checkpoint, use_fast=True)\n",
    "\n",
    "model = AutoModelForSequenceClassification.from_pretrained(\n",
    "    model_checkpoint, num_labels=NUM_CLASSES\n",
    ")\n",
    "\n",
    "model.eval()\n",
    "model.cpu()\n",
    "ds_test = ErcTextDataset(\n",
    "    DATASET=DATASET,\n",
    "    SPLIT=\"test\",\n",
    "    speaker_mode=speaker_mode,\n",
    "    num_past_utterances=num_past_utterances,\n",
    "    num_future_utterances=num_future_utterances,\n",
    "    model_checkpoint=\"roberta-large\",\n",
    "    ROOT_DIR=ROOT_DIR,\n",
    "    SEED=SEED,\n",
    ")\n",
    "\n",
    "# ds_train = ErcTextDataset(DATASET=DATASET, SPLIT='train', speaker_mode=speaker_mode,\n",
    "#                          num_past_utterances=num_past_utterances, num_future_utterances=num_future_utterances,\n",
    "#                          model_checkpoint='roberta-large',\n",
    "#                          ROOT_DIR=ROOT_DIR, SEED=SEED)\n",
    "\n",
    "# ds_val = ErcTextDataset(DATASET=DATASET, SPLIT='val', speaker_mode=speaker_mode,\n",
    "#                          num_past_utterances=num_past_utterances, num_future_utterances=num_future_utterances,\n",
    "#                          model_checkpoint='roberta-large',\n",
    "#                          ROOT_DIR=ROOT_DIR, SEED=SEED)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "ffb848c2-5839-4147-ad8e-2e665fc7d25f",
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_random_sample(ds_test, tokenizer, idx=None, max_tokens=512):\n",
    "    while True:\n",
    "        if idx is None:\n",
    "            idx_ = np.random.randint(0, len(ds_test))\n",
    "        else:\n",
    "            idx_ = idx\n",
    "        random_sample = ds_test[idx_]\n",
    "        input_ids, attention_mask, labelid = (\n",
    "            random_sample[\"input_ids\"],\n",
    "            random_sample[\"attention_mask\"],\n",
    "            random_sample[\"label\"],\n",
    "        )\n",
    "        break\n",
    "    #         if len(input_ids) < max_tokens and labelid != 0:\n",
    "    #             break\n",
    "\n",
    "    decoded = tokenizer.decode(input_ids)\n",
    "\n",
    "    input_ids = torch.tensor(input_ids).view(-1, len(input_ids))\n",
    "    attention_mask = torch.tensor(attention_mask).view(-1, len(attention_mask))\n",
    "    labelid = torch.tensor(labelid).view(-1, 1)\n",
    "\n",
    "    return idx_, input_ids, attention_mask, labelid, decoded\n",
    "\n",
    "\n",
    "def return_coeffs(\n",
    "    tokenizer,\n",
    "    input_ids,\n",
    "    attentions,\n",
    "    BATCH_IDX=0,\n",
    "    LAYER=-1,\n",
    "    QUERY_TOKEN_IDX=0,\n",
    "    annoying_char=\"Ġ\",\n",
    "):\n",
    "    tokens = tokenizer.convert_ids_to_tokens(input_ids[BATCH_IDX].tolist())\n",
    "    QUERY_TOKEN = tokens[QUERY_TOKEN_IDX].split(annoying_char)[-1]\n",
    "\n",
    "    coeffs = (\n",
    "        attentions[LAYER][BATCH_IDX].cpu().detach().numpy().sum(axis=0)[QUERY_TOKEN_IDX]\n",
    "    )\n",
    "    coeffs /= coeffs.sum()\n",
    "\n",
    "    idx_token_coeffs = [\n",
    "        (idx, token.split(annoying_char)[-1], coeffs[idx])\n",
    "        for idx, token in enumerate(tokens)\n",
    "    ]\n",
    "\n",
    "    assert len(coeffs) == len(tokens) == len(idx_token_coeffs)\n",
    "\n",
    "    return QUERY_TOKEN, coeffs, tokens, idx_token_coeffs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "id": "8ee14fb6-995f-4b88-8020-f1d78b7e18fc",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(\"<s>ELIZABETH: You're the only one I know who loves his parents. WILLIAM: I \"\n",
      " \"know.  It's going out of style isn't it? ELIZABETH: No, it's good. \"\n",
      " \"ELIZABETH: It's so lovely here.  The air is sweet. WILLIAM: So you're not \"\n",
      " \"sorry you came? ELIZABETH: No.  But I'm not going to stay. WILLIAM: Why? \"\n",
      " 'ELIZABETH: Well your mother has much as told me to leave. WILLIAM: Well-- '\n",
      " \"ELIZABETH: Well, you saw that! And you.  You've been- WILLIAM: What? \"\n",
      " 'ELIZABETH: Well sort of embarrassed soon as I came.</s></s>WILLIAM: Well the '\n",
      " 'plan was to kind of sneak up on you over a period of a week or so.  But uh- '\n",
      " \"they take it for granted that we're all set.</s>\")\n",
      "\n",
      "'data_idx: 343'\n",
      "'pred: neutral'\n",
      "'truth: happiness'\n",
      "'number of tokens in the input: 202'\n",
      "\n"
     ]
    }
   ],
   "source": [
    "import pprint\n",
    "import torch\n",
    "from tqdm.notebook import tqdm\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "idx, input_ids, attention_mask, labelid, decoded = get_random_sample(\n",
    "    ds_test, tokenizer, max_tokens=512\n",
    ")\n",
    "pprint.pprint(f\"{decoded}\")\n",
    "print()\n",
    "\n",
    "outputs = model(\n",
    "    **{\"input_ids\": input_ids, \"attention_mask\": attention_mask},\n",
    "    labels=labelid,\n",
    "    output_attentions=True,\n",
    "    output_hidden_states=True,\n",
    ")\n",
    "attentions = outputs.attentions\n",
    "pred = int(outputs.logits.argmax().numpy())\n",
    "truth = int(labelid[0][0].numpy())\n",
    "\n",
    "pprint.pprint(f\"data_idx: {idx}\")\n",
    "pprint.pprint(f\"pred: {ds_test.id2emotion[pred]}\")\n",
    "pprint.pprint(f\"truth: {ds_test.id2emotion[truth]}\")\n",
    "pprint.pprint(f\"number of tokens in the input: {input_ids.shape[1]}\")\n",
    "print()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "id": "7c6a8435-6b23-4017-9e92-fe59e7d8323d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "\"<s>ELIZABETH: You're the only one I know who loves his parents. WILLIAM: I know.  It's going out of style isn't it? ELIZABETH: No, it's good. ELIZABETH: It's so lovely here.  The air is sweet. WILLIAM: So you're not sorry you came? ELIZABETH: No.  But I'm not going to stay. WILLIAM: Why? ELIZABETH: Well your mother has much as told me to leave. WILLIAM: Well-- ELIZABETH: Well, you saw that! And you.  You've been- WILLIAM: What? ELIZABETH: Well sort of embarrassed soon as I came.</s></s>WILLIAM: Well the plan was to kind of sneak up on you over a period of a week or so.  But uh- they take it for granted that we're all set.</s>\""
      ]
     },
     "execution_count": 88,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "decoded"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "id": "4a45c106-8918-451f-a047-4a31b6f8fe0c",
   "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"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[(0, '<s>', 0.018490328), (1, 'EL', 7.087463e-05), (2, 'IZ', 0.0036216779), (3, 'AB', 9.131049e-06), (4, 'ETH', 0.003521988), (5, ':', 0.0003648079), (6, 'You', 0.0030940827), (7, \"'re\", 0.0006318671), (8, 'the', 0.012629616), (9, 'only', 2.0211244e-05), (10, 'one', 5.913559e-05), (11, 'I', 6.8649846e-05), (12, 'know', 1.880975e-05), (13, 'who', 8.237419e-05), (14, 'loves', 0.00048399763), (15, 'his', 0.00011217632), (16, 'parents', 0.0003000183), (17, '.', 0.0132984985), (18, 'WILL', 0.004431148), (19, 'I', 0.004793599), (20, 'AM', 0.012002601), (21, ':', 0.011403312), (22, 'I', 0.022841105), (23, 'know', 0.0038326092), (24, '.', 0.013298508), (25, '', 0.0054932702), (26, 'It', 0.00014509213), (27, \"'s\", 0.00083310524), (28, 'going', 8.850363e-05), (29, 'out', 3.545094e-05), (30, 'of', 3.2490403e-05), (31, 'style', 6.3238636e-05), (32, 'isn', 0.0005255854), (33, \"'t\", 0.0003637756), (34, 'it', 9.501522e-05), (35, '?', 0.003393034), (36, 'EL', 2.8172273e-05), (37, 'IZ', 0.0004412576), (38, 'AB', 4.0238824e-06), (39, 'ETH', 0.00014462559), (40, ':', 4.950233e-06), (41, 'No', 1.5734611e-05), (42, ',', 0.012476265), (43, 'it', 0.00030612876), (44, \"'s\", 1.1234338e-05), (45, 'good', 0.00032614972), (46, '.', 0.00127358), (47, 'EL', 1.5001034e-05), (48, 'IZ', 1.0666116e-05), (49, 'AB', 1.2318708e-06), (50, 'ETH', 4.3338227e-05), (51, ':', 9.699971e-07), (52, 'It', 0.00010217743), (53, \"'s\", 9.154735e-06), (54, 'so', 3.817455e-06), (55, 'lovely', 8.373584e-05), (56, 'here', 8.636514e-05), (57, '.', 0.009797152), (58, '', 2.9989222e-05), (59, 'The', 3.6566544e-06), (60, 'air', 7.055747e-07), (61, 'is', 2.3379894e-06), (62, 'sweet', 5.5980804e-06), (63, '.', 0.013298504), (64, 'WILL', 0.0033705395), (65, 'I', 0.0049620764), (66, 'AM', 0.01629891), (67, ':', 0.014080615), (68, 'So', 0.004104438), (69, 'you', 0.004694842), (70, \"'re\", 0.0035882501), (71, 'not', 0.0026207508), (72, 'sorry', 0.0036627473), (73, 'you', 0.00011639607), (74, 'came', 0.00035841804), (75, '?', 0.003645096), (76, 'EL', 2.4265279e-05), (77, 'IZ', 0.002745046), (78, 'AB', 1.5041297e-06), (79, 'ETH', 3.0804524e-05), (80, ':', 1.6235927e-05), (81, 'No', 7.5701673e-06), (82, '.', 0.013298488), (83, '', 2.4638495e-05), (84, 'But', 2.0050662e-05), (85, 'I', 4.8135206e-07), (86, \"'m\", 4.9558726e-06), (87, 'not', 2.68018e-05), (88, 'going', 7.405471e-06), (89, 'to', 0.017220613), (90, 'stay', 1.8629666e-05), (91, '.', 0.00023681644), (92, 'WILL', 0.004125756), (93, 'I', 0.003679412), (94, 'AM', 0.01670672), (95, ':', 0.014210202), (96, 'Why', 0.003255569), (97, '?', 0.0030678674), (98, 'EL', 1.3207318e-05), (99, 'IZ', 1.5809082e-05), (100, 'AB', 1.71435e-06), (101, 'ETH', 2.7766511e-05), (102, ':', 3.9893907e-06), (103, 'Well', 3.7673202e-05), (104, 'your', 0.0026127405), (105, 'mother', 0.0011114292), (106, 'has', 0.00012867739), (107, 'much', 0.00012351744), (108, 'as', 9.9839235e-05), (109, 'told', 1.6792761e-05), (110, 'me', 1.0258177e-06), (111, 'to', 1.4682961e-05), (112, 'leave', 1.5869318e-05), (113, '.', 0.0047764764), (114, 'WILL', 0.0051901652), (115, 'I', 0.004382041), (116, 'AM', 0.010670958), (117, ':', 0.015547197), (118, 'Well', 0.0038708062), (119, '--', 0.02207176), (120, 'EL', 0.000229448), (121, 'IZ', 8.87038e-06), (122, 'AB', 1.05943245e-05), (123, 'ETH', 0.00015146057), (124, ':', 2.8448203e-05), (125, 'Well', 7.6203745e-05), (126, ',', 0.012455871), (127, 'you', 6.72725e-05), (128, 'saw', 0.00014727212), (129, 'that', 0.0011304354), (130, '!', 0.000297932), (131, 'And', 0.0001657542), (132, 'you', 0.0012056122), (133, '.', 0.025552722), (134, '', 0.00024519634), (135, 'You', 0.0009340441), (136, \"'ve\", 0.0006389548), (137, 'been', 0.0037886105), (138, '-', 0.026681231), (139, 'WILL', 0.0074126534), (140, 'I', 0.005616895), (141, 'AM', 0.018437967), (142, ':', 0.034487113), (143, 'What', 0.07523814), (144, '?', 0.0076400205), (145, 'EL', 4.0721232e-05), (146, 'IZ', 4.1010903e-06), (147, 'AB', 5.049388e-06), (148, 'ETH', 1.6979962e-05), (149, ':', 3.7199297e-06), (150, 'Well', 2.14991e-05), (151, 'sort', 0.00012864286), (152, 'of', 9.370792e-05), (153, 'embarrassed', 9.636082e-05), (154, 'soon', 0.00019692183), (155, 'as', 0.00012380579), (156, 'I', 4.344883e-06), (157, 'came', 1.8656356e-05), (158, '.', 0.01329849), (159, '</s>', 0.024142722), (160, '</s>', 0.016254673), (161, 'W', 0.008152022), (162, 'ILL', 0.0041923253), (163, 'I', 0.0045123673), (164, 'AM', 0.009966147), (165, ':', 0.023438519), (166, 'Well', 0.011003701), (167, 'the', 0.010439453), (168, 'plan', 0.003520288), (169, 'was', 0.0009967433), (170, 'to', 0.0044144825), (171, 'kind', 0.000424765), (172, 'of', 0.012380396), (173, 'sneak', 0.0010113125), (174, 'up', 0.0016960267), (175, 'on', 0.0032306707), (176, 'you', 0.0011512722), (177, 'over', 0.00025279558), (178, 'a', 0.013330113), (179, 'period', 0.00014978545), (180, 'of', 0.007461519), (181, 'a', 0.016128778), (182, 'week', 0.00039316315), (183, 'or', 0.00023251718), (184, 'so', 0.00023567572), (185, '.', 0.015996318), (186, '', 0.0266955), (187, 'But', 0.013360498), (188, 'uh', 0.017351769), (189, '-', 0.015166791), (190, 'they', 0.0066828905), (191, 'take', 0.0046239137), (192, 'it', 0.024343781), (193, 'for', 0.0010041485), (194, 'granted', 0.0018707165), (195, 'that', 0.0026408685), (196, 'we', 0.01690083), (197, \"'re\", 0.00601025), (198, 'all', 0.0018313649), (199, 'set', 0.009266154), (200, '.', 0.013298489), (201, '</s>', 0.017098524)]\n",
      "\n",
      "[(143, 'What', 0.07523814), (142, ':', 0.034487113), (186, '', 0.0266955), (138, '-', 0.026681231), (133, '.', 0.025552722), (192, 'it', 0.024343781), (159, '</s>', 0.024142722), (165, ':', 0.023438519), (22, 'I', 0.022841105), (119, '--', 0.02207176)]\n"
     ]
    }
   ],
   "source": [
    "QUERY_TOKEN, coeffs, tokens, idx_token_coeffs = return_coeffs(\n",
    "    tokenizer, input_ids, attentions, LAYER=-1, QUERY_TOKEN_IDX=0\n",
    ")\n",
    "\n",
    "plt.figure()\n",
    "plt.plot(coeffs)\n",
    "plt.title(f\"query: {QUERY_TOKEN}\")\n",
    "plt.show()\n",
    "\n",
    "print(idx_token_coeffs)\n",
    "print()\n",
    "\n",
    "top_10 = sorted(idx_token_coeffs, key=lambda x: -x[2])[:10]\n",
    "print(top_10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "id": "fcad9e5b-32fe-4e51-b8a1-73081ef6c09d",
   "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"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[(0, '<s>', 0.009518345), (1, 'W', 0.0015158709), (2, 'ILL', 0.003002666), (3, 'I', 0.00079906575), (4, 'AM', 0.0015781951), (5, ':', 0.0014044514), (6, 'I', 0.0008029061), (7, 'don', 0.0012150394), (8, \"'t\", 0.0020034947), (9, 'care', 0.0015642853), (10, 'what', 0.001330332), (11, 'you', 0.0011223308), (12, 'do', 0.001359587), (13, '.', 0.0016805531), (14, '', 0.0010125607), (15, 'See', 0.0027025985), (16, ',', 0.0009157068), (17, 'you', 0.0010617579), (18, 'could', 0.0013141251), (19, 'paint', 0.0015077016), (20, 'your', 0.0008832222), (21, 'body', 0.0023187026), (22, 'bright', 0.0018949977), (23, 'green', 0.0013917314), (24, 'and', 0.0011168647), (25, 'and', 0.0011702221), (26, 'run', 0.0018754338), (27, 'naked', 0.0032058253), (28, 'through', 0.0015592488), (29, 'the', 0.00065928086), (30, 'Place', 0.0012507546), (31, 'Vend', 0.0029668831), (32, 'ome', 0.0018342098), (33, 'and', 0.0009764782), (34, 'run', 0.002030901), (35, 'with', 0.0012309302), (36, 'every', 0.0017645716), (37, 'man', 0.0017601587), (38, 'in', 0.0014209738), (39, 'the', 0.00071965816), (40, 'world', 0.0011770473), (41, 'and', 0.0010428331), (42, 'I', 0.0006280243), (43, 'sh', 0.002294852), (44, 'an', 0.0019553914), (45, \"'t\", 0.0017075946), (46, 'say', 0.0024530783), (47, 'a', 0.001143728), (48, 'word', 0.0021034176), (49, '.', 0.0015902832), (50, '', 0.0009491136), (51, 'As', 0.0014571375), (52, 'long', 0.0014925128), (53, 'as', 0.001163393), (54, 'I', 0.0006676398), (55, 'know', 0.0012178384), (56, 'that', 0.0006398568), (57, 'you', 0.0009588782), (58, 'love', 0.0019382671), (59, 'me', 0.0011169451), (60, 'best', 0.0018856209), (61, '.', 0.0012338725), (62, 'EL', 0.0070757587), (63, 'IZ', 0.003601743), (64, 'AB', 0.0018198342), (65, 'ETH', 0.003567276), (66, ':', 0.0011316314), (67, 'U', 0.0011686173), (68, 'hh', 0.002207162), (69, ',', 0.000764846), (70, 'Thank', 0.0019348701), (71, 'you', 0.000861389), (72, ',', 0.00074895285), (73, 'darling', 0.002619747), (74, '.', 0.0015790598), (75, '', 0.0009201393), (76, 'And', 0.0015987591), (77, 'the', 0.0006895473), (78, 'same', 0.0017362001), (79, 'goes', 0.002313311), (80, 'for', 0.0010366753), (81, 'you', 0.00084753573), (82, ',', 0.00069002475), (83, 'except', 0.0015788792), (84, 'that', 0.00070732954), (85, 'if', 0.0012802975), (86, 'I', 0.00060657755), (87, 'ever', 0.0014867052), (88, 'see', 0.0018439713), (89, 'you', 0.0009502966), (90, 'so', 0.0015832941), (91, 'much', 0.0008766513), (92, 'as', 0.0010658929), (93, 'look', 0.0024662088), (94, 'at', 0.0012528398), (95, 'another', 0.0008947493), (96, 'woman', 0.0011413211), (97, ',', 0.0007239629), (98, 'I', 0.00069716293), (99, \"'ll\", 0.0014978942), (100, 'kill', 0.001991496), (101, 'you', 0.00079185446), (102, '.', 0.0012827845), (103, 'WILL', 0.0031861442), (104, 'I', 0.0006697138), (105, 'AM', 0.0014575119), (106, ':', 0.0012854483), (107, 'Do', 0.0013838492), (108, 'you', 0.0008144576), (109, 'remember', 0.0019290573), (110, 'that', 0.00072831404), (111, 'awful', 0.0016432584), (112, 'scene', 0.0011812432), (113, 'we', 0.0008046751), (114, 'had', 0.0009925026), (115, 'in', 0.0010127054), (116, 'Venice', 0.0021754627), (117, '?', 0.0014124579), (118, 'EL', 0.007932011), (119, 'IZ', 0.0036833938), (120, 'AB', 0.0018959797), (121, 'ETH', 0.00375485), (122, ':', 0.0010742393), (123, 'Which', 0.0030162383), (124, 'particular', 0.001961124), (125, 'one', 0.0012009112), (126, '?', 0.0015426779), (127, 'WILL', 0.003405329), (128, 'I', 0.00078677165), (129, 'AM', 0.0013440469), (130, ':', 0.0011811582), (131, 'The', 0.0011348593), (132, 'one', 0.0012305162), (133, 'where', 0.0019920906), (134, 'you', 0.0009233365), (135, 'bought', 0.0009632544), (136, 'that', 0.00077666075), (137, 'little', 0.0015127633), (138, 'painted', 0.0016524068), (139, 'wooden', 0.0018968035), (140, 'stake', 0.0018010597), (141, 'and', 0.0008780611), (142, 'you', 0.00088610785), (143, 'put', 0.0020546787), (144, 'it', 0.0010308983), (145, 'on', 0.0009089913), (146, 'my', 0.00063739467), (147, 'bed', 0.0012369829), (148, '.', 0.0015406937), (149, 'EL', 0.008190156), (150, 'IZ', 0.003343977), (151, 'AB', 0.00204308), (152, 'ETH', 0.0034866955), (153, ':', 0.0011992406), (154, 'A', 0.001713948), (155, 'we', 0.0019653838), (156, ',', 0.0007259738), (157, 'Charles', 0.0015233135), (158, '.', 0.0016803527), (159, '', 0.0010336271), (160, 'That', 0.0013320246), (161, 'was', 0.0009149716), (162, 'his', 0.00069664547), (163, 'name', 0.0018338858), (164, '.', 0.0016604421), (165, '', 0.0010107425), (166, 'He', 0.0020922017), (167, 'did', 0.0012144969), (168, 'wr', 0.0024660383), (169, 'iggle', 0.0021544425), (170, 'so', 0.0014470869), (171, 'beautifully', 0.0020343452), (172, '.', 0.0012460385), (173, 'WILL', 0.0038329028), (174, 'I', 0.00075128436), (175, 'AM', 0.0013260451), (176, ':', 0.0009987602), (177, 'Hor', 0.002320899), (178, 'rible', 0.0023843965), (179, 'thing', 0.001217677), (180, '.', 0.0015637224), (181, '', 0.00095789036), (182, 'I', 0.00059640344), (183, 'hated', 0.00146922), (184, 'it', 0.0011707041), (185, '.', 0.0015797206), (186, 'EL', 0.012278579), (187, 'IZ', 0.0035788787), (188, 'AB', 0.0017155741), (189, 'ETH', 0.0040840833), (190, ':', 0.0012399313), (191, 'I', 0.00069241005), (192, 'know', 0.0012493262), (193, 'you', 0.0010858278), (194, 'did', 0.0010569965), (195, '.', 0.0016383964), (196, '', 0.0010527631), (197, 'You', 0.001710973), (198, 'threw', 0.001188687), (199, 'it', 0.0012651974), (200, 'out', 0.0012887444), (201, 'the', 0.00084232795), (202, 'window', 0.0014144185), (203, 'into', 0.0017886846), (204, 'the', 0.0006491534), (205, 'Canal', 0.0020272995), (206, 'below', 0.0013894619), (207, '.', 0.0013770879), (208, '', 0.0010168822), (209, 'I', 0.0005664558), (210, 'don', 0.001016838), (211, \"'t\", 0.0017285808), (212, 'think', 0.0014870327), (213, 'I', 0.00070116576), (214, \"'ll\", 0.0016081255), (215, 'ever', 0.001578364), (216, 'forgive', 0.0016252289), (217, 'you', 0.0009669229), (218, 'for', 0.0010184598), (219, 'that', 0.00080963486), (220, '.', 0.0015322631), (221, 'WILL', 0.0036032146), (222, 'I', 0.00082955917), (223, 'AM', 0.0013757685), (224, ':', 0.0012278986), (225, 'How', 0.001980457), (226, 'long', 0.0019801718), (227, 'did', 0.0013264054), (228, 'that', 0.0008498891), (229, 'rou', 0.0015672874), (230, 'e', 0.0020605526), (231, 'last', 0.0017502976), (232, '?', 0.0016724004), (233, 'EL', 0.0066304314), (234, 'IZ', 0.0032894586), (235, 'AB', 0.0019810211), (236, 'ETH', 0.0038440353), (237, ':', 0.0015206377), (238, 'It', 0.001678944), (239, 'went', 0.002162504), (240, 'on', 0.0011980484), (241, 'intermitt', 0.0022486786), (242, 'ently', 0.003414735), (243, 'for', 0.0011654384), (244, 'days', 0.0015942468), (245, '.', 0.001632802), (246, 'WILL', 0.0032867067), (247, 'I', 0.00072597165), (248, 'AM', 0.0013361525), (249, ':', 0.0014277409), (250, 'Oh', 0.0019056253), (251, ',', 0.0008067303), (252, 'The', 0.0013774916), (253, 'worst', 0.0017279934), (254, 'one', 0.0014611908), (255, 'was', 0.0010688903), (256, 'in', 0.0013491918), (257, 'Cannes', 0.002867821), (258, 'when', 0.0020267256), (259, 'your', 0.000920929), (260, 'cur', 0.0020089506), (261, 'ling', 0.0021843866), (262, 'iron', 0.0029227932), (263, 'burnt', 0.0024286339), (264, 'a', 0.0011258256), (265, 'hole', 0.0015868649), (266, 'in', 0.0014323441), (267, 'my', 0.0008881983), (268, 'new', 0.002026345), (269, 'dressing', 0.0019221322), (270, 'gown', 0.0018292363), (271, '?', 0.0020207919), (272, 'EL', 0.010533403), (273, 'IZ', 0.003677797), (274, 'AB', 0.0021084596), (275, 'ETH', 0.0044457437), (276, ':', 0.0013852217), (277, 'Oh', 0.0023030236), (278, ',', 0.00089508307), (279, 'It', 0.0018144824), (280, 'burnt', 0.0029961076), (281, 'my', 0.0008868554), (282, 'comb', 0.0018204699), (283, 'and', 0.0011120513), (284, 'all', 0.0015530415), (285, 'the', 0.00066606974), (286, 'towels', 0.0016942915), (287, 'in', 0.0015273656), (288, 'the', 0.00078108377), (289, 'bathroom', 0.00086611096), (290, '.', 0.0020294322), (291, 'WILL', 0.0044161896), (292, 'I', 0.0010784023), (293, 'AM', 0.0016620621), (294, ':', 0.0014326073), (295, '[', 0.0012724536), (296, 'LA', 0.001969536), (297, 'UGH', 0.003445215), (298, 'TER', 0.003032271), (299, ']', 0.0016274356), (300, 'That', 0.0016944726), (301, 'was', 0.0013571242), (302, 'quit', 0.0021560832), (303, 'a', 0.0014460412), (304, 'r', 0.0029968754), (305, 'ous', 0.002360298), (306, 'er', 0.00362737), (307, ',', 0.00091753295), (308, 'wasn', 0.0015089274), (309, \"'t\", 0.0020105229), (310, 'it', 0.0013186515), (311, '?', 0.0021278753), (312, 'EL', 0.007252721), (313, 'IZ', 0.0040781354), (314, 'AB', 0.00232375), (315, 'ETH', 0.004466204), (316, ':', 0.0015591467), (317, 'That', 0.0016072015), (318, 'was', 0.0011389841), (319, 'the', 0.0007495802), (320, 'first', 0.0018734161), (321, 'time', 0.002377191), (322, 'you', 0.0011044071), (323, 'hit', 0.0017353619), (324, 'me', 0.0017737106), (325, '.', 0.0022411582), (326, 'WILL', 0.004615163), (327, 'I', 0.0009885407), (328, 'AM', 0.0018061268), (329, ':', 0.0016775292), (330, 'Oh', 0.0024246909), (331, ',', 0.0010943869), (332, 'I', 0.0008424459), (333, 'didn', 0.0015612975), (334, \"'t\", 0.0024238366), (335, 'hit', 0.0017480963), (336, 'you', 0.0013640389), (337, 'very', 0.003261705), (338, 'hard', 0.0016353234), (339, '.', 0.002074804), (340, 'EL', 0.01149087), (341, 'IZ', 0.0044755023), (342, 'AB', 0.0024093466), (343, 'ETH', 0.0054314435), (344, ':', 0.00183319), (345, 'And', 0.0021084116), (346, 'then', 0.002463444), (347, 'the', 0.0008641479), (348, 'manager', 0.002783647), (349, 'came', 0.002613677), (350, 'in', 0.002271365), (351, 'and', 0.0015824672), (352, 'saw', 0.0022881147), (353, 'us', 0.0014972175), (354, 'all', 0.0022597674), (355, 'rolling', 0.0040793573), (356, 'all', 0.0022702473), (357, 'over', 0.0016074758), (358, 'the', 0.0010083106), (359, 'floor', 0.0017720651), (360, ',', 0.0013879391), (361, 'biting', 0.002067268), (362, 'and', 0.0014361786), (363, 'scratching', 0.0033824167), (364, 'one', 0.0016522728), (365, 'another', 0.0014188116), (366, '.', 0.002454297), (367, '[', 0.0012557417), (368, 'LA', 0.0023237576), (369, 'UGH', 0.0041797985), (370, 'TER', 0.0036133176), (371, ']', 0.0019690068), (372, 'WILL', 0.005371019), (373, 'I', 0.0010774347), (374, 'AM', 0.0018984943), (375, ':', 0.0019749731), (376, 'I', 0.0010364624), (377, 'should', 0.0019231569), (378, 'never', 0.0027771608), (379, 'forget', 0.0021268243), (380, 'his', 0.0013016138), (381, 'face', 0.0019633828), (382, '.', 0.0027549663), (383, '[', 0.0013604198), (384, 'LA', 0.002402352), (385, 'UGH', 0.0045240466), (386, 'TER', 0.004366175), (387, ']', 0.0022622398), (388, 'EL', 0.010032554), (389, 'IZ', 0.005355816), (390, 'AB', 0.002768981), (391, 'ETH', 0.0062624603), (392, ':', 0.001983648), (393, 'How', 0.0023858254), (394, 'ridiculous', 0.0019374024), (395, ',', 0.0015517022), (396, 'utterly', 0.002988961), (397, 'ridiculous', 0.0018578096), (398, 'WILL', 0.0063707433), (399, 'I', 0.0014741464), (400, 'AM', 0.0023587034), (401, ':', 0.0020676022), (402, 'We', 0.0027261705), (403, 'were', 0.0022702008), (404, 'very', 0.004226658), (405, 'much', 0.0018525784), (406, 'younger', 0.002794074), (407, 'then', 0.003582709), (408, '.', 0.003434017), (409, 'EL', 0.008690408), (410, 'IZ', 0.0063454113), (411, 'AB', 0.0034965225), (412, 'ETH', 0.0077680033), (413, ':', 0.002331107), (414, 'Very', 0.004215052), (415, 'much', 0.0017955189), (416, 'sill', 0.004324587), (417, 'ier', 0.0024083306), (418, '.', 0.003606934), (419, 'WILL', 0.0076448964), (420, 'I', 0.0020904785), (421, 'AM', 0.0036082063), (422, ':', 0.002547924), (423, 'Come', 0.004100004), (424, 'to', 0.0021502106), (425, 'think', 0.0029948438), (426, 'of', 0.002579485), (427, 'it', 0.0033331083), (428, ',', 0.0016556644), (429, 'the', 0.0014898345), (430, 'real', 0.0031221095), (431, 'cause', 0.004431539), (432, 'of', 0.0025469866), (433, 'that', 0.0018030746), (434, 'rou', 0.0029527617), (435, 'es', 0.0039992696), (436, 'was', 0.002192683), (437, 'Peter', 0.006909324), (438, 'B', 0.0047403206), (439, 'urden', 0.0028184731), (440, '.', 0.00474553), (441, '</s>', 0.0019689447), (442, '</s>', 0.0020101033), (443, 'EL', 0.011038411), (444, 'IZ', 0.010087603), (445, 'AB', 0.005057669), (446, 'ETH', 0.011849758), (447, ':', 0.0029198478), (448, 'You', 0.0051289005), (449, 'knew', 0.002696076), (450, 'there', 0.0031675012), (451, 'was', 0.0027643659), (452, 'nothing', 0.003421207), (453, 'in', 0.0029068943), (454, 'that', 0.0017352798), (455, '.', 0.0046727634), (456, '</s>', 0.0020223788)]\n",
      "\n",
      "[(186, 'EL', 0.012278579), (446, 'ETH', 0.011849758), (340, 'EL', 0.01149087), (443, 'EL', 0.011038411), (272, 'EL', 0.010533403), (444, 'IZ', 0.010087603), (388, 'EL', 0.010032554), (0, '<s>', 0.009518345), (409, 'EL', 0.008690408), (149, 'EL', 0.008190156)]\n"
     ]
    }
   ],
   "source": [
    "QUERY_TOKEN, coeffs, tokens, idx_token_coeffs = return_coeffs(\n",
    "    tokenizer, input_ids, attentions, LAYER=0, QUERY_TOKEN_IDX=443\n",
    ")\n",
    "\n",
    "plt.figure()\n",
    "plt.plot(coeffs)\n",
    "plt.title(f\"query: {QUERY_TOKEN}\")\n",
    "plt.show()\n",
    "\n",
    "print(idx_token_coeffs)\n",
    "print()\n",
    "\n",
    "top_10 = sorted(idx_token_coeffs, key=lambda x: -x[2])[:10]\n",
    "print(top_10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a8ed6951-dc77-4c20-9e5d-01b060bffd27",
   "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": "3.7.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}