-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
184 lines (153 loc) · 6.15 KB
/
main.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
import torch
import torch.nn as nn
import torch.optim as optim
import copy
import os
from torchtext import data
from utils import load_data
from sst_sent import SST_SENT
from recurrent_models import RNN_encoder
from slstm import SLSTM
from IPython import embed
from tensorboardX import SummaryWriter
def do_forward_pass(batch, s_encoder, loss_function):
l_probs, h_l, attention_weights = s_encoder(
batch.text[0], batch.text[1])
# l_probs, h_l, attention_weights = s_encoder(
# batch.text[0])
# calculate loss
loss = loss_function(l_probs, batch.label - 1)
# calculate accuracy
_, predictions = torch.max(l_probs.data, 1)
k_ = config['k_']
topk_classes = torch.topk(l_probs.data, k_)[1] + 1
filter_ = torch.eq(topk_classes, batch.label.data.unsqueeze(1))
# embed()
acc = float(torch.sum(torch.eq(predictions, batch.label.data - 1))) \
/ predictions.size()[0]
acc_k = float(torch.sum(filter_)) / predictions.size()[0]
return acc, loss, h_l, acc_k
# embed()
writer = SummaryWriter()
writer_path = list(writer.all_writers.keys())[0]
best_model_path = os.path.join(writer_path, 'best_dev_model')
if not os.path.exists(best_model_path):
os.makedirs(best_model_path)
# train, dev, test, inputs, answers = load_data(SST_SENT, 'SST_FINE_PHRASES')
# train, dev, test, inputs, answers = load_data(SST_SENT, 'SST_FINE')
train, dev, test, inputs, answers = load_data(SST_SENT, 'SST_BIN')
# train, dev, test, inputs, answers = load_data(SST_SENT, 'SST_BIN_PHRASES')
if torch.cuda.is_available():
device_ = 0
else:
device_ = -1
# load data and word embeddings
train_iter, dev_iter, test_iter = data.BucketIterator.splits(
(train, dev, test),
batch_sizes=(100, len(dev.examples), len(test.examples)),
sort_key=lambda x: len(x.text), device=device_, sort_within_batch=True)
train_iter.init_epoch()
dev_iter.init_epoch()
test_iter.init_epoch()
# dnn_encoder = RNN_encoder(
# input_dim=inputs.vocab.vectors.size()[1],
# output_dim=100,
# num_classes=len(answers.vocab.freqs.keys()),
# vocab=inputs.vocab)
dnn_encoder = SLSTM(
input_dim=inputs.vocab.vectors.size()[1],
hidden_size=100,
num_layers=1,
window=2,
num_classes=len(answers.vocab.freqs.keys()),
vocab=inputs.vocab)
if torch.cuda.is_available():
dnn_encoder.cuda(0)
# define loss funtion and optimizer
loss_function = nn.NLLLoss()
optimizer = optim.Adam(dnn_encoder.parameters())
# optimizer = optim.Adam(dnn_encoder.parameters(), weight_decay=1e-5)
save_graph_of_model = True
train_losses_list = []
train_acc_list = []
dev_acc_list = []
dev_losses_list = []
dev_max_acc = 0
max_acc_k = 0
acc_step = 0
best_model = None
config = {}
config['epochs'] = 5
config['k_'] = 2
wd = 1e-3 # weight decay
print('Starting training procedure')
# start training procedure
for batch_idx, batch in enumerate(train_iter):
# embed()
# switch to to training mode,zero out gradients
dnn_encoder.train()
optimizer.zero_grad()
acc, loss, _, acc_k = do_forward_pass(
batch, dnn_encoder, loss_function)
loss.backward()
# embed()
# weight regularization
for group in optimizer.param_groups:
for param in group['params']:
param.data = param.data.add(-wd * group['lr'], param.data)
optimizer.step()
train_acc_list.append(acc)
train_losses_list.append(float(loss))
# embed()
writer.add_scalar('train/Loss', float(loss), batch_idx)
writer.add_scalar('train/Acc', acc, batch_idx)
writer.add_scalar('timers_train/bf_after', float(dnn_encoder.bef_aft_time_el), batch_idx)
writer.add_scalar('timers_train/sent_time_el', float(dnn_encoder.sent_time_el), batch_idx)
writer.add_scalar('timers_train/words_time_gates_el', float(dnn_encoder.words_time_gates_el), batch_idx)
writer.add_scalar('timers_train/words_time_rest_el', float(dnn_encoder.words_time_rest_el), batch_idx)
# evaluate on dev set
# with torch.no_grad():
dnn_encoder.eval()
dev_batch = next(iter(dev_iter))
# acc, loss, _ = perform_forward_pass(dev_batch, dnn_model, loss_function)
acc, loss, _, acc_k = do_forward_pass(
dev_batch, dnn_encoder, loss_function)
dev_acc_list.append(acc)
dev_losses_list.append(float(loss))
writer.add_scalar('timers_dev/bf_after', float(dnn_encoder.bef_aft_time_el), batch_idx)
writer.add_scalar('timers_dev/sent_time_el', float(dnn_encoder.sent_time_el), batch_idx)
writer.add_scalar('timers_dev/words_time_gates_el', float(dnn_encoder.words_time_gates_el), batch_idx)
writer.add_scalar('timers_dev/words_time_rest_el', float(dnn_encoder.words_time_rest_el), batch_idx)
if acc > dev_max_acc:
dev_max_acc = acc
best_model = copy.deepcopy(dnn_encoder)
max_acc_k = acc_k
acc_step = batch_idx
# check acc on test set
test_batch = next(iter(test_iter))
t_acc, t_loss, _, acc_k = do_forward_pass(test_batch, dnn_encoder, loss_function)
print('accuraccy on test set is {} and at topk {}'.format(
t_acc, acc_k))
writer.add_scalar('dev/Loss', float(loss), batch_idx)
writer.add_scalar('dev/Acc', acc, batch_idx)
# getting tracing of legacy functions not supported error
if save_graph_of_model:
# with SummaryWriter(comment='Net') as w:
# w.add_graph(dnn_encoder, (dev_batch.text[0], ))
# writer.add_graph(dnn_encoder, batch.text[0])
# l_probs, h_l, attention_weights = dnn_encoder(batch.text[0])
# embed()
# error check
# https://github.com/lanpa/tensorboard-pytorch/pull/106
# writer.add_graph(dnn_encoder, (batch.text[0], batch.text[1]), verbose=True)
# torch.onnx.export(dnn_encoder, (batch.text[0], batch.text[1]), "./IndexLayer.pb", verbose=True)
save_graph_of_model = False
# info and stop criteria
if train_iter.iterations % 100 == 0:
print('epoch {} iteration {} current train acc {} train loss {} max dev acc {} at k {} at step {} \n'.format(
train_iter.epoch, batch_idx, train_acc_list[-1],
train_losses_list[-1], dev_max_acc, max_acc_k, acc_step))
if train_iter.epoch > config['epochs']:
break
writer.close()
embed()