-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathclassification.py
205 lines (165 loc) · 7.91 KB
/
classification.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
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
#!/usr/bin/env python
# -*- coding: utf-8 -*-
__author__ = "Han"
__email__ = "liuhan132@foxmail.com"
"""Multi-Label Classification"""
import json
import argparse
from tqdm import tqdm
import torch
import torch.nn
import torch.multiprocessing
import logging
from models import *
from datareaders import *
from utils.optims import Optim
from utils.metrics import *
from utils.config import init_logging, init_env
logger = logging.getLogger(__name__)
def main(config_path, in_infix, out_infix, is_train, is_test, gpuid):
logger.info('-------------Multi-Label Classification---------------')
logger.info('initial environment...')
config, enable_cuda, device, writer = init_env(config_path, in_infix, out_infix,
writer_suffix='cls_log_path', gpuid=gpuid)
logger.info('reading dataset...')
# dataset = DocRepClsReader(config)
dataset = DocClsReader(config)
logger.info('constructing model...')
# model = MultiLabelCls(config).to(device)
model = E2EMultiLabelCls(config).to(device)
model.load_parameters(enable_cuda) # replace=('tar_doc_encoder', 'encoder')
# loss function
criterion = MultiLabelCls.criterion
optimizer = Optim(config['train']['optimizer'],
lr=config['train']['learning_rate'],
max_grad_norm=config['train']['clip_grad_norm'],
lr_decay=config['train']['learning_rate_decay'],
start_decay_at=config['train']['start_decay_at'])
optimizer.set_parameters(model.parameters())
# dataset
train_data = dataset.get_dataloader_train()
valid_data = dataset.get_dataloader_valid()
test_data = dataset.get_dataloader_test()
if is_train:
logger.info('start training...')
num_epochs = config['train']['num_epochs']
best_metrics = None
best_epoch = 0
for epoch in range(1, num_epochs + 1):
# train
model.train() # set training = True, make sure right dropout
train_on_model(epoch=epoch,
model=model,
criterion=criterion,
optimizer=optimizer,
batch_data=train_data,
device=device,
writer=writer)
optimizer.updateLearningRate(epoch) # learning rate decay
# evaluate
with torch.no_grad():
model.eval() # let training = False, make sure right dropout
metrics, _ = eval_on_model(model=model,
batch_data=valid_data,
device=device)
logger.info('epoch={}: '.format(epoch) + print_metrics(metrics, stage='valid'))
# save best model with maximum micro-f1 and macro-f1
if best_metrics is None or metrics['micro_f1'] + metrics['macro_f1'] > \
best_metrics['micro_f1'] + best_metrics['macro_f1']:
model.save_parameters(epoch)
best_metrics = metrics
best_epoch = epoch
logging.info('best epoch={}: '.format(best_epoch) + print_metrics(best_metrics, stage='valid'))
with open('outputs/' + out_infix + '/valid_metrics.json', 'w') as wf:
json.dump(best_metrics, wf, indent=2)
if is_test:
logger.info('start testing...')
if is_train:
model.load_out_parameters(enable_cuda, force=True, strict=True)
with torch.no_grad():
model.eval()
metrics, predict = eval_on_model(model=model,
batch_data=test_data,
device=device)
logger.info(print_metrics(metrics, stage='test'))
with open('outputs/' + out_infix + '/metrics.json', 'w') as wf:
json.dump(metrics, wf, indent=2)
torch.save(predict, 'outputs/' + out_infix + '/predict.pt')
writer.close()
logger.info('finished.')
def print_metrics(metrics, stage='test'):
out = "\n{stage}_macro_f1={macro_f1:.2%}," \
"\n{stage}_micro_f1={micro_f1:.2%}," \
"\n{stage}_micro_p={micro_p:.2%}," \
"\n{stage}_micro_r={micro_r:.2%}," \
"\n{stage}_hamming_loss={hl:.4f}," \
"\n{stage}_one_error={oe:.2%}" \
.format(stage=stage, macro_f1=metrics['macro_f1'], micro_f1=metrics['micro_f1'],
micro_p=metrics['micro_p'], micro_r=metrics['micro_r'],
hl=metrics['hamming_loss'], oe=metrics['one_error'])
return out
def train_on_model(epoch, model, criterion, optimizer, batch_data, device, writer):
batch_cnt = len(batch_data)
sum_loss = 0.
for i, batch in tqdm(enumerate(batch_data), total=batch_cnt, desc='Training Epoch=%d' % epoch):
model.zero_grad()
# batch data
batch = [x.to(device) if x is not None else x for x in batch]
truth = batch[-1]
batch_input = batch[:-1]
# forward
predict = model.forward(*batch_input)
loss = criterion(predict, truth)
loss.backward()
# evaluate
macro_f1, micro_f1, micro_p, micro_r, _ = evaluate_f1_ml(predict, truth)
hamming_loss = evaluate_hamming_loss(predict, truth)
one_error = evaluate_one_error(predict, truth)
optimizer.step() # update parameters
# logging
batch_loss = loss.item()
sum_loss += batch_loss * truth.shape[0]
writer.add_scalar('Train-Step-Loss', batch_loss, global_step=epoch * batch_cnt + i)
writer.add_scalar('Train-Step-Macro_F1', macro_f1, global_step=epoch * batch_cnt + i)
writer.add_scalar('Train-Step-Micro_F1', micro_f1, global_step=epoch * batch_cnt + i)
writer.add_scalar('Train-Step-Micro_P', micro_p, global_step=epoch * batch_cnt + i)
writer.add_scalar('Train-Step-Micro_R', micro_r, global_step=epoch * batch_cnt + i)
writer.add_scalar('Train-Step-Hamming_Loss', hamming_loss, global_step=epoch * batch_cnt + i)
writer.add_scalar('Train-Step-One_Error', one_error, global_step=epoch * batch_cnt + i)
def eval_on_model(model, batch_data, device):
batch_cnt = len(batch_data)
all_predict = []
all_truth = []
for i, batch in tqdm(enumerate(batch_data), total=batch_cnt, desc='Testing...'):
# batch data
batch = [x.to(device) if x is not None else x for x in batch]
truth = batch[-1]
all_truth.append(truth)
batch_input = batch[:-1]
# forward
predict = model.forward(*batch_input)
all_predict.append(predict)
predict = torch.cat(all_predict, dim=0)
truth = torch.cat(all_truth, dim=0)
macro_f1, micro_f1, micro_p, micro_r, label_f1 = evaluate_f1_ml(predict, truth)
hamming_loss = evaluate_hamming_loss(predict, truth)
one_error = evaluate_one_error(predict, truth)
metrics = {'macro_f1': macro_f1,
'micro_f1': micro_f1,
'micro_p': micro_p,
'micro_r': micro_r,
'hamming_loss': hamming_loss,
'one_error': one_error,
'label_f1': label_f1}
return metrics, predict
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('-config', type=str, default='config.yaml', help='config path')
parser.add_argument('-in', dest='in_infix', type=str, default='default', help='input data_path infix')
parser.add_argument('-out', type=str, default='default', help='output data_path infix')
parser.add_argument('-train', action='store_true', default=False, help='enable train step')
parser.add_argument('-test', action='store_true', default=False, help='enable test step')
parser.add_argument('-gpuid', type=int, default=None, help='gpuid')
args = parser.parse_args()
init_logging(out_infix=args.out)
main(args.config, args.in_infix, args.out, is_train=args.train, is_test=args.test, gpuid=args.gpuid)