-
Notifications
You must be signed in to change notification settings - Fork 40
/
stanford-skip.py
173 lines (140 loc) · 7.13 KB
/
stanford-skip.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
# -*- coding: utf-8 -*-
# author: Tiancheng Zhao
from __future__ import print_function
import argparse
import json
import logging
import os
import torch
from laed import evaluators, utt_utils, dialog_utils
from laed import main as engine
from laed.dataset import corpora
from laed.dataset import data_loaders
from laed.models import dialog_models
from laed.utils import str2bool, prepare_dirs_loggers, get_time, process_config
arg_lists = []
parser = argparse.ArgumentParser()
def add_argument_group(name):
arg = parser.add_argument_group(name)
arg_lists.append(arg)
return arg
def get_config():
config, unparsed = parser.parse_known_args()
return config, unparsed
# Data
data_arg = add_argument_group('Data')
data_arg.add_argument('--data_dir', type=str, nargs='+', default=['data/stanford'])
data_arg.add_argument('--log_dir', type=str, default='logs')
# Network
net_arg = add_argument_group('Network')
net_arg.add_argument('--use_reg_kl', type=str2bool, default=True)
net_arg.add_argument('--use_attribute', type=str2bool, default=True)
net_arg.add_argument('--y_size', type=int, default=3)
net_arg.add_argument('--k', type=int, default=5)
net_arg.add_argument('--rnn_cell', type=str, default='gru')
net_arg.add_argument('--embed_size', type=int, default=200)
net_arg.add_argument('--utt_type', type=str, default='rnn')
net_arg.add_argument('--utt_cell_size', type=int, default=256)
net_arg.add_argument('--ctx_cell_size', type=int, default=512)
net_arg.add_argument('--dec_cell_size', type=int, default=512)
net_arg.add_argument('--bi_ctx_cell', type=str2bool, default=False)
net_arg.add_argument('--max_utt_len', type=int, default=40)
net_arg.add_argument('--max_dec_len', type=int, default=40)
net_arg.add_argument('--max_vocab_cnt', type=int, default=10000)
net_arg.add_argument('--num_layer', type=int, default=1)
net_arg.add_argument('--use_attn', type=str2bool, default=False)
net_arg.add_argument('--attn_type', type=str, default='cat')
net_arg.add_argument('--greedy_q', type=str2bool, default=True)
# Training / test parameters
train_arg = add_argument_group('Training')
train_arg.add_argument('--op', type=str, default='adam')
train_arg.add_argument('--backward_size', type=int, default=10)
train_arg.add_argument('--step_size', type=int, default=1)
train_arg.add_argument('--grad_clip', type=float, default=3.0)
train_arg.add_argument('--init_w', type=float, default=0.08)
train_arg.add_argument('--init_lr', type=float, default=0.001)
train_arg.add_argument('--momentum', type=float, default=0.0)
train_arg.add_argument('--lr_hold', type=int, default=1)
train_arg.add_argument('--lr_decay', type=float, default=0.6)
train_arg.add_argument('--dropout', type=float, default=0.3)
train_arg.add_argument('--improve_threshold', type=float, default=0.996)
train_arg.add_argument('--patient_increase', type=float, default=2.0)
train_arg.add_argument('--early_stop', type=str2bool, default=True)
train_arg.add_argument('--max_epoch', type=int, default=50)
train_arg.add_argument('--loss_type', type=str, default="e2e")
train_arg.add_argument('--include_eod', type=str2bool, default=True)
# MISC
misc_arg = add_argument_group('Misc')
misc_arg.add_argument('--save_model', type=str2bool, default=True)
misc_arg.add_argument('--use_gpu', type=str2bool, default=True)
misc_arg.add_argument('--fix_batch', type=str2bool, default=False)
misc_arg.add_argument('--print_step', type=int, default=200)
misc_arg.add_argument('--ckpt_step', type=int, default=1000)
misc_arg.add_argument('--freeze_step', type=int, default=5000)
misc_arg.add_argument('--batch_size', type=int, default=30)
misc_arg.add_argument('--preview_batch_num', type=int, default=1)
misc_arg.add_argument('--gen_type', type=str, default='greedy')
misc_arg.add_argument('--avg_type', type=str, default='word')
misc_arg.add_argument('--beam_size', type=int, default=10)
misc_arg.add_argument('--forward_only', type=str2bool, default=False)
data_arg.add_argument('--load_sess', type=str, default="2018-02-14T12-55-34-stanford-skip.py")
logger = logging.getLogger()
def main(config):
prepare_dirs_loggers(config, os.path.basename(__file__))
corpus_client = corpora.StanfordCorpus(config)
dial_corpus = corpus_client.get_corpus()
train_dial, valid_dial, test_dial = dial_corpus['train'],\
dial_corpus['valid'],\
dial_corpus['test']
evaluator = evaluators.BleuEvaluator("CornellMovie")
# create data loader that feed the deep models
train_feed = data_loaders.SMDDialogSkipLoader("Train", train_dial, config)
valid_feed = data_loaders.SMDDialogSkipLoader("Valid", valid_dial, config)
test_feed = data_loaders.SMDDialogSkipLoader("Test", test_dial, config)
model = dialog_models.StED(corpus_client, config)
if config.forward_only:
test_file = os.path.join(config.log_dir, config.load_sess,
"{}-test-{}.txt".format(get_time(), config.gen_type))
dump_file = os.path.join(config.log_dir, config.load_sess,
"{}-z.pkl".format(get_time()))
model_file = os.path.join(config.log_dir, config.load_sess, "model")
else:
test_file = os.path.join(config.session_dir,
"{}-test-{}.txt".format(get_time(), config.gen_type))
dump_file = os.path.join(config.session_dir, "{}-z.pkl".format(get_time()))
model_file = os.path.join(config.session_dir, "model")
if config.use_gpu:
model.cuda()
if config.forward_only is False:
try:
engine.train(model, train_feed, valid_feed,
test_feed, config, evaluator, gen=dialog_utils.generate_with_adv)
except KeyboardInterrupt:
print("Training stopped by keyboard.")
config.batch_size = 10
model.load_state_dict(torch.load(model_file))
engine.validate(model, valid_feed, config)
engine.validate(model, test_feed, config)
dialog_utils.generate_with_adv(model, test_feed, config, None, num_batch=0)
selected_clusters = utt_utils.latent_cluster(model, train_feed, config, num_batch=None)
selected_outs = dialog_utils.selective_generate(model, test_feed, config, selected_clusters)
print(len(selected_outs))
with open(os.path.join(dump_file+'.json'), 'wb') as f:
json.dump(selected_clusters, f, indent=2)
with open(os.path.join(dump_file+'.out.json'), 'wb') as f:
json.dump(selected_outs, f, indent=2)
with open(os.path.join(dump_file), "wb") as f:
print("Dumping test to {}".format(dump_file))
dialog_utils.dump_latent(model, test_feed, config, f, num_batch=None)
with open(os.path.join(test_file), "wb") as f:
print("Saving test to {}".format(test_file))
dialog_utils.gen_with_cond(model, test_feed, config, num_batch=None,
dest_f=f)
with open(os.path.join(test_file+'.txt'), "wb") as f:
print("Saving test to {}".format(test_file))
dialog_utils.generate(model, test_feed, config, evaluator, num_batch=None,
dest_f=f)
if __name__ == "__main__":
config, unparsed = get_config()
config = process_config(config)
main(config)