-
Notifications
You must be signed in to change notification settings - Fork 5
/
train_classifier.py
150 lines (123 loc) · 6.07 KB
/
train_classifier.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import model_classifier
import argparse
import datasets
from ignite.engine import Events, create_supervised_trainer, create_supervised_evaluator
from ignite.metrics import CategoricalAccuracy, Loss
from ignite.handlers import EarlyStopping
from torch.utils.data import DataLoader
import os
import json
def main():
parser = argparse.ArgumentParser(description='Training')
parser.add_argument('--learning_rate', type=float, default=0.0001,
help='Output filename')
parser.add_argument('--batch_size', type=int, default=32,
help='Output filename')
parser.add_argument('--epochs', type=int, default=200,
help='Epochs')
parser.add_argument('--dataset', type=str, default="Names",
help='Output filename')
parser.add_argument('--checkpoints_directory', type=str, default="CKPTS",
help='Check Points Directory')
parser.add_argument('--hidden_units', type=int, default=256,
help='hidden_units')
parser.add_argument('--embedding_size', type=int, default=256,
help='embedding_size')
parser.add_argument('--patience', type=int, default=10,
help='patience')
parser.add_argument('--classifier_type', type=str, default="charRNN",
help='rnn type')
args = parser.parse_args()
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
train_dataset = datasets.get_dataset(args.dataset, dataset_type = 'train')
val_dataset = datasets.get_dataset(args.dataset, dataset_type = 'train_val')
if args.classifier_type == "charRNN":
model_options = {
'vocab_size' : len(train_dataset.idx_to_char),
'hidden_size' : args.hidden_units,
'target_size' : len(train_dataset.classes),
'embedding_size' : args.embedding_size
}
model = model_classifier.uniRNN(model_options)
print "char RNN"
if args.classifier_type == "biRNN":
model_options = {
'vocab_size' : len(train_dataset.idx_to_char),
'hidden_size' : args.hidden_units,
'target_size' : len(train_dataset.classes),
'embedding_size' : args.embedding_size
}
model = model_classifier.biRNN(model_options)
print "BI RNN"
if args.classifier_type == "CNN":
model_options = {
'vocab_size' : len(train_dataset.idx_to_char),
'hidden_size' : args.hidden_units,
'target_size' : len(train_dataset.classes),
'embedding_size' : args.embedding_size
}
model = model_classifier.CnnTextClassifier(model_options)
print "CnnTextClassifier"
print device
model.to(device)
parameters = filter(lambda p: p.requires_grad, model.parameters())
optimizer = optim.Adam(parameters, lr=args.learning_rate)
loss_criterion = nn.CrossEntropyLoss()
print "check", torch.cuda.is_available()
train_loader = DataLoader(train_dataset, batch_size=args.batch_size,
shuffle=True, num_workers=0)
val_loader = DataLoader(val_dataset, batch_size=args.batch_size,
shuffle=True, num_workers=0)
trainer = create_supervised_trainer(model, optimizer, loss_criterion)
evaluator = create_supervised_evaluator(model,
metrics={
'accuracy': CategoricalAccuracy(),
'nll': Loss(loss_criterion)
})
checkpoints_dir = "{}/{}_classifer_{}".format(args.checkpoints_directory, args.dataset, args.classifier_type)
if not os.path.exists(checkpoints_dir):
os.makedirs(checkpoints_dir)
training_log = {
'model_options' : model_options,
'log' : [],
'best_epoch' : 0,
'best_accuracy' : 0.0
}
@trainer.on(Events.ITERATION_COMPLETED)
def log_training_loss(trainer):
total_batches = int(len(train_dataset)/args.batch_size)
if trainer.state.iteration % 100 == 0:
print("Epoch[{}] Iteration[{}] Total Iterations[{}] Loss: {:.2f}".format(
trainer.state.epoch, trainer.state.iteration, total_batches, trainer.state.output))
@trainer.on(Events.EPOCH_COMPLETED)
def log_validation_results(trainer):
evaluator.run(train_loader)
training_metrics = evaluator.state.metrics
print("Training Results - Epoch: {} Avg accuracy: {:.2f} Avg loss: {:.2f}"
.format(trainer.state.epoch, training_metrics['accuracy'], training_metrics['nll']))
evaluator.run(val_loader)
evaluation_metrics = evaluator.state.metrics
print("Validation Results - Epoch: {} Avg accuracy: {:.2f} Avg loss: {:.2f}"
.format(trainer.state.epoch, evaluation_metrics['accuracy'], evaluation_metrics['nll']))
out_path = "{}/model_epoch_{}.pth".format(checkpoints_dir, trainer.state.epoch)
torch.save(model.state_dict(), out_path)
training_log['log'].append({
'training_metrics' : training_metrics,
'evaluation_metrics' : evaluation_metrics,
})
# if (trainer.state.epoch - training_log['best_epoch']) > args.patience and (evaluation_metrics['accuracy'] < training_log['best_accuracy']):
# trainer.terminate()
if evaluation_metrics['accuracy'] > training_log['best_accuracy']:
torch.save(model.state_dict(), "{}/best_model.pth".format(checkpoints_dir))
training_log['best_accuracy'] = evaluation_metrics['accuracy']
training_log['best_epoch'] = trainer.state.epoch
print "BEST", training_log['best_epoch'], training_log['best_accuracy']
with open("{}/training_log.json".format(checkpoints_dir), 'w') as f:
f.write(json.dumps(training_log))
trainer.run(train_loader, max_epochs=args.epochs)
if __name__ == '__main__':
main()