-
Notifications
You must be signed in to change notification settings - Fork 13
/
evaluate.py
312 lines (249 loc) · 12.6 KB
/
evaluate.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
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
import argparse
import os
import random
import torch
from torch import nn
from torch.nn import functional as F
import numpy as np
import math
from str2bool import str2bool
from datetime import datetime
from utils import (
save_hparams,
KGDataset,
collate_fn,
get_batch_loader,
DisBatcher,
GenBatcher,
)
from metrics import (
bleu_metric,
distinct_metric,
f1_metric
)
from model.extract import PtrExtractSumm
from model.rl import PolicyGradient
from model.util import sequence_loss, weighted_sequence_loss
from transformers import GPT2PreTrainedModel, GPT2Model, GPT2Config
class GPT2Summ(GPT2PreTrainedModel):
'''succeed from GPT2PreTraninedModel which has implemented the 'generate' func'''
def __init__(self, tokenizer, gpt2_config, segment=True):
config = GPT2Config.from_pretrained(gpt2_config)
super(GPT2Summ, self).__init__(config)
self.transformer = GPT2Model.from_pretrained(gpt2_config)
self.transformer.resize_token_embeddings(len(tokenizer))
self.user_id = [tokenizer.convert_tokens_to_ids('<user1>'),
tokenizer.convert_tokens_to_ids('<user2>')]
self.know_id = tokenizer.convert_tokens_to_ids('<knowledge>')
self.segment = segment
self.lm_head = nn.Linear(config.n_embd, len(tokenizer), bias=False)
self.config.vocab_size = len(tokenizer)
self.tie_weights()
def get_output_embeddings(self):
return self.lm_head
def prepare_inputs_for_generation(self, input_ids, **kwargs):
token_type_ids = []
for i in range(input_ids.size(0)):
ids = input_ids[i].tolist()
type_ids = []
last_special_token = self.know_id
for j in range(len(ids)):
if ids[j] in ([self.know_id] + self.user_id):
type_ids.append(ids[j])
last_special_token = ids[j]
else:
type_ids.append(last_special_token)
token_type_ids.append(type_ids)
token_type_ids = torch.tensor(token_type_ids).type_as(input_ids)
# only last token for inputs_ids if past is defined in kwargs
if "past" in kwargs and kwargs["past"]:
input_ids = input_ids[:, -1].unsqueeze(-1)
token_type_ids = token_type_ids[:, -1].unsqueeze(-1)
if self.segment:
inputs = {"input_ids": input_ids, "token_type_ids": token_type_ids}
else:
inputs = {"input_ids": input_ids}
inputs.update(kwargs)
return inputs
def forward(self, input_ids, past=None, attention_mask=None, token_type_ids=None):
transformer_outputs = self.transformer(input_ids, past=past, token_type_ids=token_type_ids)
hidden_states = transformer_outputs[0]
lm_logits = self.lm_head(hidden_states)
return (lm_logits,) + transformer_outputs[1:]
def batch_decode(self, input_ids, max_len, min_len, early_stopping, beam_size,
repetition_penalty, eos_id, length_penalty, no_repeat_ngram_size):
# new-version
output_sequences = self.generate(
input_ids=input_ids,
max_length=input_ids.size(1) + max_len,
min_length=input_ids.size(1) + min_len,
do_sample=False,
early_stopping=early_stopping,
num_beams=beam_size,
repetition_penalty=repetition_penalty,
pad_token_id=0,
# pad_token_id=None,
eos_token_id=eos_id,
length_penalty=length_penalty,
no_repeat_ngram_size=no_repeat_ngram_size,
)
return output_sequences
def load_gen_net(tokenizer, segment, gpt2_config, gen_pretrain_file, load=True, cuda=True):
gen = GPT2Summ(tokenizer=tokenizer, gpt2_config=gpt2_config, segment=segment)
if load:
print("Restoring all non-adagrad variables from {}...".format(gen_pretrain_file))
state_dict = torch.load(gen_pretrain_file)['state_dict']
gen.load_state_dict(state_dict)
if cuda:
gen = gen.cuda()
return gen
def load_dis_net(emb_dim, lstm_hidden, lstm_layer, bert_config, dis_pretrain_file, load=True, cuda=True):
dis = PtrExtractSumm(
emb_dim=emb_dim, lstm_hidden=lstm_hidden, lstm_layer=lstm_layer, bert_config=bert_config)
dis = PolicyGradient(dis.transformer, dis._extractor)
if load:
print("Restoring all non-adagrad variables from {}...".format(dis_pretrain_file))
state_dict = torch.load(dis_pretrain_file)['state_dict']
dis.load_state_dict(state_dict)
if cuda:
dis = dis.cuda()
return dis
def main(args):
print("\nParameters:")
for attr, value in sorted(vars(args).items()):
print("{}={}".format(attr.upper(), value))
print("")
# Selecting wihch GPU to use
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu_list
args.cuda = torch.cuda.is_available() and not args.no_cuda
# Output directory for models and summaries
out_dir = os.path.join(args.log, args.exp_name)
if not os.path.exists(out_dir):
os.makedirs(out_dir)
print('Writing to {}\n'.format(out_dir))
save_hparams(args, os.path.join(out_dir, 'hparams'))
# Checkpoint directory
checkpoint_dir = os.path.join(out_dir, 'checkpoints')
checkpoint_prefix = os.path.join(checkpoint_dir, 'model')
if not os.path.exists(checkpoint_dir):
os.makedirs(checkpoint_dir)
# Build dataset
time_str = datetime.now().strftime('%Y-%m-%d %H:%M:%S')
print("Create training dataset begain... | %s " % time_str)
test_seen_dataset = KGDataset(args.test_seen_file, max_knowledge=999)
test_unseen_dataset = KGDataset(args.test_unseen_file, max_knowledge=999)
test_seen_loader = get_batch_loader(test_seen_dataset, collate_fn=collate_fn, batch_size=args.eval_batch_size, is_test=True)
test_unseen_loader = get_batch_loader(test_unseen_dataset, collate_fn=collate_fn, batch_size=args.eval_batch_size, is_test=True)
time_str = datetime.now().strftime('%Y-%m-%d %H:%M:%S')
print("Create training dataset end... | %s " % time_str)
# Batcher
dis_batcher = DisBatcher(args.bert_truncate, args.bert_config, args.cuda)
gen_batcher = GenBatcher(args.knowledge_truncate, args.text_truncate, args.gpt2_truncate, args.gpt2_config, args.cuda)
# Load model
dis_model = load_dis_net(args.emb_dim, args.lstm_hidden, args.lstm_layer, args.bert_config, args.dis_pretrain_file, args.load_dis, args.cuda)
gen_model = load_gen_net(gen_batcher.tokenizer, args.segment, args.gpt2_config, args.gen_pretrain_file, args.load_gen, args.cuda)
ce = lambda logit, target: F.cross_entropy(logit, target, reduce=False)
gen_criterion = lambda logits, targets: sequence_loss(logits, targets, ce, pad_idx=-1)
def dev_step(split, global_step):
if split == 'test_seen':
test_loader = test_seen_loader
elif split == 'test_unseen':
test_loader = test_unseen_loader
else:
raise ValueError
dis_model.eval()
gen_model.eval()
n_token, test_loss = 0, 0.0 # ppl
test_hyp, test_ref = [], []
count = 0
with torch.no_grad():
for knowledges, histories, users, responses, knowledge_lens in test_loader:
knowledges = [know.split('\n\n') for know in knowledges]
histories = [his.split('\n\n') for his in histories]
dis_args = dis_batcher(knowledges, histories, knowledge_lens, args.n_sent)
dis_out = dis_model(*dis_args)
dis_knowledges = [[knowledges[bi][dis_out[0][bi].item()]] for bi in range(len(knowledges))]
gen_args = gen_batcher(dis_knowledges, histories, users, responses, args.segment, True)
loss = gen_criterion(gen_model(gen_args[0], token_type_ids=gen_args[1])[0], gen_args[2])
n_token += loss.size(0)
test_loss += loss.sum().item()
for bi in range(len(dis_knowledges)):
dec_in = gen_batcher(dis_knowledges[bi:bi+1], histories[bi:bi+1], users[bi:bi+1], segment=args.segment, training=False)
dec_out = gen_model.batch_decode(dec_in, args.max_length, args.min_length, args.early_stopping,
args.beam_size, args.repetition_penalty, gen_batcher.eos_id,
args.length_penalty, args.no_repeat_ngram_size)
dec_out = dec_out[0].tolist()[dec_in.size(1):]
_hyp = gen_batcher.tokenizer.decode(dec_out, skip_special_tokens=True, clean_up_tokenization_spaces=False)
_ref = responses[bi]
test_hyp.append(_hyp)
test_ref.append(_ref)
count += 1
if count % 1000 == 0:
print(count)
with open(os.path.join(out_dir, '{}-decoded-iter-{}.txt'.format(split, global_step)), 'w') as f:
for _hyp, _ref in zip(test_hyp, test_ref):
f.writelines('{} ||| {}\n'.format(_hyp, _ref))
MeanLoss = test_loss / n_token
b1, b2, b3, b4 = bleu_metric(test_hyp, test_ref)
d1, d2 = distinct_metric(test_hyp)
f1 = f1_metric(test_hyp, test_ref)
time_str = datetime.now().strftime('%Y-%m-%d %H:%M:%S')
print("**********************************")
print("{} results..........".format(split))
print('hypothesis: ', len(test_hyp))
print("Step: %d \t| ppl: %.3f \t| %s" % (global_step, math.exp(MeanLoss), time_str))
print("BLEU-1/2/3/4: {:.4f}/{:.4f}/{:.4f}/{:.4f}".format(b1, b2, b3, b4))
print("Distinct-1/2: {:.4f}/{:.4f}".format(d1, d2))
print("F1: {:.4f}".format(f1))
print("**********************************")
return {'f1': f1, 'loss': MeanLoss, 'bleu1': b1, 'bleu2': b2, 'bleu3': b3, 'bleu4': b4, 'distinct1': d1, 'distinct2': d2}
dev_step("test_seen", 0) # test_random_split
dev_step("test_unseen", 0) # test_topic_split
if __name__ == '__main__':
parser = argparse.ArgumentParser(
description='Pre-training for Knowledge-Grounded Conversation'
)
# files
parser.add_argument('--test_seen_file', type=str, default='wizard_of_wikipedia/data/test_seen.jsonl')
parser.add_argument('--test_unseen_file', type=str, default='wizard_of_wikipedia/data/test_unseen.jsonl')
# training scheme
parser.add_argument('--eval_batch_size', type=int, default=1)
# save
parser.add_argument('--exp_name', type=str, default='test')
parser.add_argument('--log', type=str, default='wizard_of_wikipedia/log')
parser.add_argument('--seed', type=int, default=42)
# pre-train
parser.add_argument('--dis_pretrain_file', type=str, default='wizard_of_wikipedia/checkpoints/model-dis-best')
parser.add_argument('--gen_pretrain_file', type=str, default='wizard_of_wikipedia/checkpoints/model-gen-best')
parser.add_argument('--load_dis', type=str2bool, default=True)
parser.add_argument('--load_gen', type=str2bool, default=True)
# model
parser.add_argument('--bert_config', type=str, default='pretrain-models/bert_base_uncased')
parser.add_argument('--gpt2_config', type=str, default='pretrain-models/gpt2')
parser.add_argument('--bert_truncate', type=int, default=64) # for bert
parser.add_argument('--gpt2_truncate', type=int, default=256) # for gpt2
parser.add_argument('--knowledge_truncate', type=int, default=64) # for gpt2
parser.add_argument('--text_truncate', type=int, default=64) # for gpt2
parser.add_argument('--segment', type=str2bool, default=True)
parser.add_argument('--n_sent', type=int, default=1)
parser.add_argument('--max_length', type=int, default=30)
parser.add_argument('--min_length', type=int, default=15)
parser.add_argument('--early_stopping', type=str2bool, default=False)
parser.add_argument('--beam_size', type=int, default=1)
parser.add_argument('--repetition_penalty', type=float, default=1.0)
parser.add_argument('--length_penalty', type=float, default=1.0)
parser.add_argument('--no_repeat_ngram_size', type=int, default=0)
parser.add_argument('--emb_dim', type=int, default=768)
parser.add_argument('--lstm_hidden', type=int, default=256)
parser.add_argument('--lstm_layer', type=int, default=1)
# gpu
parser.add_argument('--gpu_list', type=str, default='2')
parser.add_argument('--gpu_ratio', type=float, default=0.85)
parser.add_argument('--no_cuda', type=str2bool, default=False)
args = parser.parse_args()
torch.manual_seed(args.seed)
torch.cuda.manual_seed_all(args.seed)
np.random.seed(args.seed)
random.seed(args.seed)
torch.backends.cudnn.deterministic = True
main(args)