-
Notifications
You must be signed in to change notification settings - Fork 0
/
decode_baseline_cand.py
280 lines (229 loc) · 11.1 KB
/
decode_baseline_cand.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
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
""" run decoding of X-ext (+ abs)"""
import argparse
import json
import os
from os.path import join
from datetime import timedelta
from time import time
from collections import Counter, defaultdict
from itertools import product
from functools import reduce
import operator as op
import heapq
from cytoolz import identity, concat, curry
import torch
from torch.utils.data import DataLoader
from torch import multiprocessing as mp
from data.batcher import tokenize
from decoding import Abstractor, Extractor, DecodeDataset, BeamAbstractor
from decoding import make_html_safe
MAX_ABS_NUM = 6 # need to set max sentences to extract for non-RL extractor
def decode(save_path, abs_dir, ext_dir, split, batch_size, max_len, num_candidates, beam_size, diverse,
final_rerank, keep_original_sent, cuda, abstracted, debug=False):
start = time()
# setup model
assert beam_size >= num_candidates and num_candidates > 1
if abstracted:
abstractor = identity
else:
if beam_size == 1:
abstractor = Abstractor(abs_dir, max_len, cuda)
else:
abstractor = BeamAbstractor(abs_dir, max_len, cuda)
extractor = Extractor(ext_dir, max_ext=MAX_ABS_NUM, cuda=cuda)
emb_type = extractor.emb_type
print("emb_type")
print(emb_type)
# setup loader
def coll(batch):
articles = list(filter(bool, batch))
return articles
dataset = DecodeDataset(split)
n_data = len(dataset)
loader = DataLoader(
dataset, batch_size=batch_size, shuffle=False, num_workers=4,
collate_fn=coll
)
# prepare save paths and logs
for i in range(MAX_ABS_NUM):
os.makedirs(join(save_path, 'output_{}'.format(i)))
dec_log = {}
dec_log['abstractor'] = json.load(open(join(abs_dir, 'meta.json')))
dec_log['extractor'] = json.load(open(join(ext_dir, 'meta.json')))
dec_log['rl'] = False
dec_log['split'] = split
dec_log['beam'] = beam_size
with open(join(save_path, 'log.json'), 'w') as f:
json.dump(dec_log, f, indent=4)
# Decoding
i = 0
with torch.no_grad():
for i_debug, raw_article_batch in enumerate(loader):
#tokenized_article_batch = list(map(tokenize(None), raw_article_batch))
tokenized_article_batch = list(map(tokenize(None, emb_type, num_candidates), raw_article_batch))
art_ids = []
if abstracted:
num_passed_sents = 0
for art_sents in tokenized_article_batch:
art_ids += [(num_passed_sents, len(art_sents))]
num_passed_sents += len(art_sents)
else:
tokenized_article_batch_flattened = [] # a list of tokenized sentence for all the articles in the batch
for art_sents in tokenized_article_batch:
art_ids += [(len(tokenized_article_batch_flattened), len(art_sents))]
tokenized_article_batch_flattened += art_sents
if beam_size > 1:
all_beams = abstractor(tokenized_article_batch_flattened, beam_size, diverse) # a list of beam for the whole batch
dec_outs = rerank_mp(all_beams, art_ids, num_candidates - 1, final_rerank=final_rerank)
# dec_outs: a list of list of token list [total number of sentences in batch, num_candidates, seq_len]
else:
dec_outs = abstractor(tokenized_article_batch_flattened)
tokenized_article_batch_flattened = []
assert i == batch_size * i_debug
if debug:
print("dec_outs[0]")
print(dec_outs[0])
print("dec_outs[1]")
print(dec_outs[1])
print("length of dec_out")
print(len(dec_outs))
print("article output")
for batch_i, (j, n) in enumerate(art_ids):
# one article
raw_article_sents = raw_article_batch[batch_i]
if abstracted:
art_sents_with_cands = tokenized_article_batch[batch_i]
else:
art_sents_with_cands = [] # a list of tokenized sentence candidates for one article
for sent_i, sent in enumerate(dec_outs[j:j + n]):
# one sent
if beam_size > 1:
candidate_list = sent
else:
candidate_list = [sent]
if keep_original_sent:
candidate_list.insert(0, tokenized_article_batch[batch_i][sent_i])
art_sents_with_cands += candidate_list
if debug:
print("art_sents_with_cands: {}".format(' '.join(art_sents_with_cands[0])))
print("art_sents_with_cands: {}".format(' '.join(art_sents_with_cands[1])))
print("art_sents_with_cands: {}".format(' '.join(art_sents_with_cands[2])))
print("art_sents_with_cands: {}".format(' '.join(art_sents_with_cands[3])))
# extraction
ext = extractor(art_sents_with_cands)
if debug:
print("ext: {}".format(ext))
# write to .dec
for k, ext_i in enumerate(ext):
#ext_str = ' '.join(art_sents_with_cands[ext_i])
ext_str = raw_article_sents[ext_i]
if debug:
print("k: {}, i: {}, sent: {}".format(k, i, ext_str))
with open(join(save_path, 'output_{}/{}.dec'.format(k, i)),
'w') as f:
f.write(make_html_safe(ext_str))
i += 1
print('{}/{} ({:.2f}%) decoded in {} seconds\r'.format(
i, n_data, i / n_data * 100, timedelta(seconds=int(time() - start))
), end='')
if debug:
raise ValueError
print()
_PRUNE = defaultdict(
lambda: 2,
{1:5, 2:5, 3:5, 4:5, 5:5, 6:4, 7:3, 8:3}
)
def rerank(all_beams, ext_inds, k, final_rerank=False):
beam_lists = (all_beams[i: i+n] for i, n in ext_inds if n > 0)
# a list of beam list, each beam list contains the beam for one article
if final_rerank:
topked = map(rereank_topk_one(k=k), beam_lists)
else:
topked = map(topk_one(k=k), beam_lists)
return list(concat(topked))
def rerank_mp(all_beams, ext_inds, k, final_rerank=False):
beam_lists = [all_beams[i: i+n] for i, n in ext_inds if n > 0]
# a list of beam list, each beam list contains the beam for one article
with mp.Pool(8) as pool:
if final_rerank:
topked = pool.map(rereank_topk_one(k=k), beam_lists)
else:
topked = pool.map(topk_one(k=k), beam_lists)
return list(concat(topked)) # a list contains the candidates sentences for all articles in the batch
@curry
def rereank_topk_one(beams, k):
"""
:param beams: a list of beam in one article
:param k:
:return: art_dec_outs: a list of list of token list, len(art_dec_outs)=num_sents_in_article, len(art_dec_outs[0])=num_cands_in_sent_0
"""
@curry
def process_beam(beam, n):
for b in beam[:n]:
b.gram_cnt = Counter(_make_n_gram(b.sequence))
return beam[:n]
beams = map(process_beam(n=_PRUNE[len(beams)]), beams)
beams_with_topk_hyps = [heapq.nlargest(k, hyps, key=_compute_score) for hyps in beams]
art_dec_outs = []
for topk_hyps in beams_with_topk_hyps:
art_dec_outs.append([h.sequence for h in topk_hyps])
return art_dec_outs
@curry
def topk_one(beams, k):
# beams: a list of beam in one article
art_dec_outs = [] # a list of token list for an article, each token list is a candidate sentence
for hyps in beams: # hypotheses for each input sentence
sent_candidates = [h.sequence for h in hyps[:k]]
art_dec_outs.append(sent_candidates)
return art_dec_outs
def _make_n_gram(sequence, n=2):
return (tuple(sequence[i:i+n]) for i in range(len(sequence)-(n-1)))
def _compute_score(hyp):
repeat = sum(c-1 for g, c in hyp.gram_cnt.items() if c > 1)
lp = hyp.logprob / len(hyp.sequence)
return (-repeat, lp)
if __name__ == '__main__':
parser = argparse.ArgumentParser(
description=('combine an extractor and an abstractor '
'to decode summary and evaluate on the '
'CNN/Daily Mail dataset')
)
parser.add_argument('--path', required=True, help='path to store/eval')
parser.add_argument('--abs_dir', help='root of the abstractor model', required=True)
parser.add_argument('--ext_dir', help='root of the extractor model', required=True)
# dataset split
#data = parser.add_mutually_exclusive_group(required=True)
#data.add_argument('--val', action='store_true', help='use validation set')
#data.add_argument('--test', action='store_true', help='use test set')
# decode options
parser.add_argument('--batch', type=int, action='store', default=32,
help='batch size of faster decoding')
parser.add_argument('--n_ext', type=int, action='store', default=4,
help='number of sents to be extracted')
parser.add_argument('--max_dec_word', type=int, action='store', default=30,
help='maximun words to be decoded for the abstractor')
parser.add_argument('--no-cuda', action='store_true',
help='disable GPU training')
parser.add_argument('--beam', type=int, action='store', default=1,
help='beam size of abstractor decoding (reranking included)')
parser.add_argument('--div', type=float, action='store', default=1.0,
help='diverse ratio for the diverse beam-search')
parser.add_argument('--num_candidates', type=int, action='store', default=2,
help='The number of candidates for each sentence.')
parser.add_argument('--final_rerank', action='store_true',
help='rereank the output of diverse beam search by n-gram repeat')
parser.add_argument('--remove_original_sentence', action='store_true',
help='remove the original sentence from the candidates')
parser.add_argument('--test_set_folder', type=str, action='store', default="test",
help='The name of testing set folder')
parser.add_argument('--abstracted', action='store_true',
help='Inidcate the test set already been abstracted, so the agent will not do the abstraction anymore')
parser.add_argument('--debug', action='store_true',
help='debug')
args = parser.parse_args()
args.cuda = torch.cuda.is_available() and not args.no_cuda
keep_original_sentence = not args.remove_original_sentence
#data_split = 'test' if args.test else 'val'
decode(args.path, args.abs_dir, args.ext_dir,
args.test_set_folder, args.batch, args.max_dec_word, args.num_candidates, args.beam, args.div,
args.final_rerank, keep_original_sentence, args.cuda, args.abstracted, args.debug)