-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
199 lines (173 loc) · 8.02 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
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
# coding: utf-8
import argparse
import math
import os
import random
import time
from typing import List
import torch.optim as optim
from tqdm import tqdm
import global_variables
from io_module.data_loader import *
from io_module.logger import get_logger, change_handler
from model.LM import *
LOGGER = None
# perplexity calculator
def evaluate(dataset, model, device, batch_size=10, ntokens=10):
model.eval()
total_loss = 0.0
total_length = 0
with torch.no_grad():
for j in tqdm(range(math.ceil(len(dataset) / batch_size))):
samples = dataset[j * batch_size: (j + 1) * batch_size]
batch_samples, masks = standardize_batch(samples, ntokens)
batch_length = torch.sum(masks).item()
total_loss += model.get_loss(batch_samples.to(device), masks.to(device)).item() * batch_length
total_length += batch_length
return total_loss / (total_length - 1)
# out is [batch, max_len+2]
def standardize_batch(sentence_list: List, ntokens=10) -> (torch.Tensor, torch.Tensor):
max_len = max([len(sentence) for sentence in sentence_list])
standardized_list = []
mask_list = []
for sentence in sentence_list:
standardized_sentence = [ntokens] + sentence + [ntokens + 1] + [0] * (max_len - len(sentence))
mask = [1] * (len(sentence) + 2) + [0] * (max_len - len(sentence))
standardized_list.append(np.array(standardized_sentence))
mask_list.append(mask)
return torch.from_numpy(np.array(standardized_list)).long(), torch.from_numpy(np.array(mask_list)).long()
def main():
parser = argparse.ArgumentParser(description="Gaussian Input Output HMM")
parser.add_argument(
'--data',
type=str,
default='./dataset/syntic_data_yong/0-1000-10-new/',
help='location of the data corpus')
parser.add_argument('--epoch', type=int, default=50)
parser.add_argument('--batch', type=int, default=10)
parser.add_argument('--lr', type=float, default=0.001)
parser.add_argument('--momentum', type=float, default=0.9)
parser.add_argument('--var_scale', type=float, default=1.0)
parser.add_argument('--log_dir', type=str,
default='./output/' + datetime.datetime.now().strftime("%Y-%m-%d_%H%M%S") + "/")
parser.add_argument('--dim', type=int, default=10)
parser.add_argument('--gpu', action='store_true')
parser.add_argument('--random_seed', type=int, default=10)
args = parser.parse_args()
# np.random.seed(global_variables.RANDOM_SEED)
# torch.manual_seed(global_variables.RANDOM_SEED)
# random.seed(global_variables.RANDOM_SEED)
np.random.seed(args.random_seed)
torch.manual_seed(args.random_seed)
random.seed(args.random_seed)
epoch = args.epoch
batch_size = args.batch
lr = args.lr
momentum = args.momentum
root = args.data
# TODO ntokens generate from dataset
ntokens = 1000
# save parameter
logger = get_logger('IOHMM', log_dir=args.log_dir)
# logger = LOGGER
logger.info(args)
logger.info('Parameter From global_variables.py')
logger.info('LOG_PATH:' + global_variables.LOG_PATH)
logger.info('EMISSION_CHO_GRAD:' + str(global_variables.EMISSION_CHO_GRAD))
logger.info('TRANSITION_CHO_GRAD:' + str(global_variables.TRANSITION_CHO_GRAD))
logger.info('DECODE_CHO_GRAD:' + str(global_variables.DECODE_CHO_GRAD))
logger.info('FAR_TRANSITION_MU:' + str(global_variables.FAR_TRANSITION_MU))
logger.info('FAR_DECODE_MU:' + str(global_variables.FAR_DECODE_MU))
logger.info('FAR_EMISSION_MU:' + str(global_variables.FAR_EMISSION_MU))
# TODO temp
logger.info('RANDOM_SEED:' + str(args.random_seed))
device = torch.device('cuda') if args.gpu else torch.device('cpu')
# Loading data
logger.info('Load data....')
train_dataset = hmm_generate_data_loader(root, type='train')
dev_dataset = hmm_generate_data_loader(root, type='dev')
test_dataset = hmm_generate_data_loader(root, type='test')
# build model
logger.info("Building model....")
# model = GaussianBatchLanguageModel(dim=args.dim, ntokens=ntokens)
model = MixtureGaussianBatchLanguageModel(dim=args.dim, ntokens=ntokens)
# model = RNNLanguageModel("LSTM", ntokens=ntokens, ninp=10, nhid=10)
model.to(device)
optimizer = optim.Adam(model.parameters(), lr=lr)
train_ppl_list = []
dev_ppl_list = []
test_ppl_list = []
# depend on dev ppl
best_epoch = 0
for i in range(epoch):
epoch_loss = 0.0
total_length = 0
random.shuffle(train_dataset)
model.train()
optimizer.zero_grad()
for j in tqdm(range(math.ceil(len(train_dataset) / batch_size))):
samples = train_dataset[j * batch_size: (j + 1) * batch_size]
input_sample, mask = standardize_batch(samples)
batch_length = torch.sum(mask).item()
total_length += batch_length
loss = model.get_loss(input_sample.to(device), mask.to(device))
loss.backward()
optimizer.step()
optimizer.zero_grad()
epoch_loss += loss.item() * batch_length
epoch_loss = epoch_loss / (total_length - 1)
time.sleep(0.5)
logger.info("Epoch:\t" + str(i) + "\t Training loss:\t" + str(round(epoch_loss, 4)) + "\t PPL: " + str(
round(np.exp(epoch_loss), 4)))
# evaluate
dev_loss = evaluate(dev_dataset, model, device, ntokens=ntokens)
test_loss = evaluate(test_dataset, model, device, ntokens=ntokens)
time.sleep(0.5)
logger.info("\t\t Dev Loss: " + str(round(dev_loss, 4)) + "\t PPL: " + str(round(np.exp(dev_loss), 4)))
logger.info("\t\t Test Loss: " + str(round(test_loss, 4)) + "\t PPL: " + str(round(np.exp(test_loss), 4)))
train_ppl_list.append(np.exp(epoch_loss))
dev_ppl_list.append(np.exp(dev_loss))
test_ppl_list.append(np.exp(test_loss))
if np.exp(dev_loss) < dev_ppl_list[best_epoch]:
best_epoch = i
total_dev, masks = standardize_batch(dev_dataset, ntokens=ntokens)
predict, corr_cnt, corr_acc = model.inference(total_dev, masks)
logger.info("\t\t Dev Correct Number " + str(corr_cnt) + "\t Correct Acc: " + str(round(corr_acc, 4)))
logger.info('=' * 10 + ' best result ' + '=' * 10)
logger.info(
'Epoch: ' +
str(best_epoch) +
'\tTrain PPL: ' +
str(round(train_ppl_list[best_epoch], 4)) +
'\tDev PPL: ' +
str(round(dev_ppl_list[best_epoch], 4)) +
'\tTest PPL: ' +
str(round(test_ppl_list[best_epoch], 4)))
# TODO it is ugly. Maybe DFS is a good method
def grid_search():
global LOGGER
for emission_cho_grad in (False, True):
global_variables.EMISSION_CHO_GRAD = emission_cho_grad
for transition_cho_grad in (False, True):
global_variables.TRANSITION_CHO_GRAD = transition_cho_grad
for decode_cho_grad in (False, True):
global_variables.DECODE_CHO_GRAD = decode_cho_grad
for far_emission_mu in (False, True):
global_variables.FAR_EMISSION_MU = far_emission_mu
for far_transition_mu in (False, True):
global_variables.FAR_TRANSITION_MU = far_transition_mu
for far_decode_mu in (False, True):
global_variables.FAR_DECODE_MU = far_decode_mu
global_variables.LOG_PATH = './output/' + datetime.datetime.now().strftime(
"%Y-%m-%d_%H%M%S") + "/"
if not os.path.exists(global_variables.LOG_PATH):
os.makedirs(global_variables.LOG_PATH)
if LOGGER is None:
LOGGER = get_logger('IOHMM')
change_handler(LOGGER, global_variables.LOG_PATH)
try:
main()
except:
LOGGER.warning("Error in this setting!")
if __name__ == '__main__':
main()