-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathgenerate_laed_features.py
167 lines (137 loc) · 6.81 KB
/
generate_laed_features.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
from __future__ import print_function
import argparse
import logging
import os
import sys
import numpy as np
import pickle
sys.path.append(os.path.join(os.path.dirname(__file__), 'NeuralDialog-LAED'))
sys.path.append(os.path.join(os.path.dirname(__file__), 'NeuralDialog-ZSDG'))
from laed.dataset import data_loaders
from laed.utils import str2bool, prepare_dirs_loggers, get_time, process_config
from laed.enc2dec.decoders import TEACH_FORCE
from utils import corpora
from utils.corpora import load_vocab, load_model, load_config
arg_lists = []
parser = argparse.ArgumentParser()
logger = logging.getLogger()
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
def process_data_feed(model, feed, config):
features = []
model.eval()
feed.epoch_init(config, shuffle=False, verbose=True)
while True:
batch = feed.next_batch()
if batch is None:
break
laed_out = model.forward(batch, TEACH_FORCE, config.gen_type, return_latent=True)
laed_z = laed_out['y_ids']
features.append(laed_z.data.cpu().numpy())
return np.array(features).reshape(-1, config.y_size * config.k)
def deflatten_laed_features(in_laed_features, in_dialogs, pad_mode=None):
pad = np.zeros_like(in_laed_features[0])
result = []
start_turn = 0
for dialog_i in in_dialogs:
if pad_mode == 'start_end':
dialog_i_turns = np.concatenate([[pad],
in_laed_features[start_turn: start_turn + len(dialog_i) - 2,:],
[pad]],
axis=0)
start_turn += len(dialog_i) - 2
elif pad_mode == 'start':
dialog_i_turns = np.concatenate([[pad],
in_laed_features[start_turn: start_turn + len(dialog_i) - 1,:]],
axis=0)
start_turn += len(dialog_i) - 1
else:
dialog_i_turns = in_laed_features[start_turn: start_turn + len(dialog_i),:]
start_turn += len(dialog_i)
while len(dialog_i) < dialog_i_turns.shape[0]:
dialog_i_turns = np.delete(dialog_i_turns, (-1), axis=0)
result.append(dialog_i_turns)
assert len(in_dialogs) == len(result)
return result
def main(config):
laed_config = load_config(config.model)
laed_config.use_gpu = config.use_gpu
laed_config = process_config(laed_config)
setattr(laed_config, 'black_domains', config.black_domains)
setattr(laed_config, 'black_ratio', config.black_ratio)
setattr(laed_config, 'include_domain', True)
setattr(laed_config, 'include_example', False)
setattr(laed_config, 'include_state', True)
setattr(laed_config, 'entities_file', 'NeuralDialog-ZSDG/data/stanford/kvret_entities.json')
setattr(laed_config, 'action_match', True)
setattr(laed_config, 'batch_size', config.batch_size)
setattr(laed_config, 'data_dir', config.data_dir)
setattr(laed_config, 'include_eod', False) # for StED model
setattr(laed_config, 'domain_description', config.domain_description)
if config.process_seed_data:
assert config.corpus_client[:3] == 'Zsl', 'Incompatible coprus_client for --process_seed_data flag'
corpus_client = getattr(corpora, config.corpus_client)(laed_config)
if config.vocab:
corpus_client.vocab, corpus_client.rev_vocab, corpus_client.unk_token = load_vocab(config.vocab)
prepare_dirs_loggers(config, os.path.basename(__file__))
dial_corpus = corpus_client.get_corpus()
# train_dial, valid_dial, test_dial = dial_corpus['train'], dial_corpus['valid'], dial_corpus['test']
# all_dial = train_dial + valid_dial + test_dial
# all_utts = reduce(lambda x, y: x + y, all_dial, [])
model = load_model(config.model, config.model_name, config.model_type, corpus_client=corpus_client)
if config.use_gpu:
model.cuda()
for dataset_name in ['train', 'valid', 'test']:
dataset = dial_corpus[dataset_name]
feed_data = dataset if config.model_type == 'dialog' else reduce(lambda x, y: x + y, dataset, [])
# create data loader that feed the deep models
if config.process_seed_data:
seed_utts = corpus_client.get_seed_responses(utt_cnt=len(corpus_client.domain_descriptions))
main_feed = getattr(data_loaders, config.data_loader)("Test", feed_data, laed_config)
features = process_data_feed(model, main_feed, laed_config)
if config.data_loader == 'SMDDialogSkipLoader':
pad_mode = 'start_end'
elif config.data_loader == 'SMDDataLoader':
pad_mode = 'start'
else:
pad_mode = None
features = deflatten_laed_features(features, dataset, pad_mode=pad_mode)
assert sum(map(len, dataset)) == sum(map(lambda x: x.shape[0], features))
if not os.path.exists(config.out_folder):
os.makedirs(config.out_folder)
with open(os.path.join(config.out_folder, 'dialogs_{}.pkl'.format(dataset_name)), 'w') as result_out:
pickle.dump(features, result_out)
if config.process_seed_data:
seed_utts = corpus_client.get_seed_responses(utt_cnt=len(corpus_client.domain_descriptions))
seed_feed = data_loaders.PTBDataLoader("Seed", seed_utts, laed_config)
seed_features = process_data_feed(model, seed_feed, laed_config)
with open(os.path.join(config.out_folder, 'seed_utts.pkl'), 'w') as result_out:
pickle.dump(seed_features, result_out)
if __name__ == "__main__":
# Data
data_arg = add_argument_group('Data')
data_arg.add_argument('model')
data_arg.add_argument('out_folder')
data_arg.add_argument('--model_name', required=True)
data_arg.add_argument('--model_type', required=True, help='sent/dialog')
data_arg.add_argument('--data_dir', nargs='+')
data_arg.add_argument('--corpus_client', required=True)
data_arg.add_argument('--data_loader', required=True, help='PTBDataLoader/SMDDataLoader/SMDDialogSkipLoader')
data_arg.add_argument('--black_domains', nargs='*', default=[])
data_arg.add_argument('--black_ratio', type=float, default=1.0)
data_arg.add_argument('--batch_size', default=1)
data_arg.add_argument('--process_seed_data', default=False, action='store_true')
data_arg.add_argument('--vocab', default=None)
data_arg.add_argument('--domain_description', default='annotation')
# MISC
misc_arg = add_argument_group('Misc')
misc_arg.add_argument('--use_gpu', type=str2bool, default=True)
misc_arg.add_argument('--forward_only', type=str2bool, default=True)
misc_arg.add_argument('--gen_type', type=str, default='greedy')
config, unparsed = get_config()
main(config)