remove pyinstrument. (dmlc#2772)

Co-authored-by: xiang song(charlie.song) <[email protected]>

examples/pytorch/graphsage/experimental/README.md

@@ -4,7 +4,6 @@ This is an example of training GraphSage in a distributed fashion. Before traini
 
 ```bash
 sudo pip3 install ogb
-sudo pip3 install pyinstrument
 ```
 
 To train GraphSage, it has five steps:

examples/pytorch/graphsage/experimental/train_dist.py

@@ -20,7 +20,6 @@
 import torch.optim as optim
 import torch.multiprocessing as mp
 from torch.utils.data import DataLoader
-from pyinstrument import Profiler
 
 def load_subtensor(g, seeds, input_nodes, device):
     """
@@ -186,9 +185,6 @@ def run(args, device, data):
 
     # Training loop
     iter_tput = []
-    profiler = Profiler()
-    if args.close_profiler == False:
-        profiler.start()
     epoch = 0
     for epoch in range(args.num_epochs):
         tic = time.time()
@@ -253,9 +249,6 @@ def run(args, device, data):
                                          g.ndata['labels'], val_nid, test_nid, args.batch_size_eval, device)
             print('Part {}, Val Acc {:.4f}, Test Acc {:.4f}, time: {:.4f}'.format(g.rank(), val_acc, test_acc,
                                                                                   time.time() - start))
-    if args.close_profiler == False:
-        profiler.stop()
-        print(profiler.output_text(unicode=True, color=True))
 
 def main(args):
     dgl.distributed.initialize(args.ip_config, args.num_servers, num_workers=args.num_workers)
@@ -313,7 +306,6 @@ def main(args):
         help="Number of sampling processes. Use 0 for no extra process.")
     parser.add_argument('--local_rank', type=int, help='get rank of the process')
     parser.add_argument('--standalone', action='store_true', help='run in the standalone mode')
-    parser.add_argument('--close_profiler', action='store_true', help='Close pyinstrument profiler')
     args = parser.parse_args()
     assert args.num_workers == int(os.environ.get('DGL_NUM_SAMPLER')), \
     'The num_workers should be the same value with DGL_NUM_SAMPLER.'

examples/pytorch/graphsage/experimental/train_dist_unsupervised.py

@@ -21,7 +21,6 @@
 import torch.optim as optim
 import torch.multiprocessing as mp
 from dgl.distributed import DistDataLoader
-#from pyinstrument import Profiler
 
 class SAGE(nn.Module):
     def __init__(self,
@@ -328,8 +327,6 @@ def run(args, device, data):
     optimizer = optim.Adam(model.parameters(), lr=args.lr)
 
     # Training loop
-    #profiler = Profiler()
-    #profiler.start()
     epoch = 0
     for epoch in range(args.num_epochs):
         sample_time = 0

examples/pytorch/rgcn/experimental/README.md

@@ -5,7 +5,7 @@ This is an example of training RGCN node classification in a distributed fashion
 Before training, install python libs by pip:
 
 ```bash
-pip3 install ogb pyinstrument pyarrow
+pip3 install ogb pyarrow
 ```
 
 To train RGCN, it has four steps:

examples/pytorch/rgcn/experimental/entity_classify_dist.py

@@ -29,7 +29,6 @@
 import tqdm
 
 from ogb.nodeproppred import DglNodePropPredDataset
-from pyinstrument import Profiler
 
 class EntityClassify(nn.Module):
     """ Entity classification class for RGCN

examples/pytorch/rgcn/experimental/get_mag_data.py

@@ -4,8 +4,6 @@
 import numpy as np
 from ogb.nodeproppred import DglNodePropPredDataset
 
-from pyinstrument import Profiler
-
 # Load OGB-MAG.
 dataset = DglNodePropPredDataset(name='ogbn-mag')
 hg_orig, labels = dataset[0]

examples/pytorch/rgcn/experimental/write_mag.py

@@ -3,7 +3,6 @@
 import torch as th
 import numpy as np
 from ogb.nodeproppred import DglNodePropPredDataset
-from pyinstrument import Profiler
 
 # Load OGB-MAG.
 dataset = DglNodePropPredDataset(name='ogbn-mag')
@@ -21,9 +20,6 @@
 #    subg_nodes[ntype] = np.random.choice(hg.number_of_nodes(ntype), int(hg.number_of_nodes(ntype) / 5), replace=False)
 #hg = dgl.compact_graphs(dgl.node_subgraph(hg, subg_nodes))
 
-profiler = Profiler()
-profiler.start()
-
 # OGB-MAG is stored in heterogeneous format. We need to convert it into homogeneous format.
 g = dgl.to_homogeneous(hg)
 g.ndata['orig_id'] = g.ndata[dgl.NID]
@@ -85,6 +81,3 @@
     eid_ranges[etype] = [int(eid[0]), int(eid[-1] + 1)]
 with open('mag.json', 'w') as outfile:
     json.dump({'nid': nid_ranges, 'eid': eid_ranges}, outfile, indent=4)
-
-profiler.stop()
-print(profiler.output_text(unicode=True, color=True))

tools/convert_partition.py

@@ -7,7 +7,6 @@
 import torch as th
 import pyarrow
 from pyarrow import csv
-from pyinstrument import Profiler
 
 parser = argparse.ArgumentParser(description='Construct graph partitions')
 parser.add_argument('--input-dir', required=True, type=str,
@@ -58,9 +57,6 @@
 etypes = [e[0] for e in etypes]
 etypes_map = {e:i for i, e in enumerate(etypes)}
 
-profiler = Profiler()
-profiler.start()
-
 def read_feats(file_name):
     attrs = csv.read_csv(file_name, read_options=pyarrow.csv.ReadOptions(autogenerate_column_names=True),
                          parse_options=pyarrow.csv.ParseOptions(delimiter=' '))
@@ -227,6 +223,3 @@ def read_feats(file_name):
                                                 'part_graph': part_graph_file}
 with open('{}/{}.json'.format(output_dir, graph_name), 'w') as outfile:
     json.dump(part_metadata, outfile, sort_keys=True, indent=4)
-
-profiler.stop()
-print(profiler.output_text(unicode=True, color=True))