-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathwoz.py
234 lines (209 loc) · 9.75 KB
/
woz.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
from transformers import GPT2Tokenizer, GPT2Model, GPT2LMHeadModel
from datasets import load_metric
from pprint import pprint
import torch
import math
import time
import sys
import json
import numpy as np
def make_woz_datasets(bKnowledge):
if bKnowledge:
out_names = ['woz.train_c.txt','woz.valid_c.txt','woz.test_c.txt']
else:
out_names = ['woz.train_b.txt','woz.valid_a.txt','woz.test_a.txt','woz.valid_b.txt','woz.test_b.txt']
max_ins = [18,2,2,2,2]
count = 0
counts = []
for dataset in range(len(out_names)):
fout = open(out_names[dataset],'wt')
for dialog in range(1,max_ins[dataset],1):
file_name = 'dialogues_%03d.json' % dialog
print(file_name)
with open(file_name) as f:
data = json.load(f)
for dialogue in data:
if len(dialogue['services']) == 1:
if dialogue['services'][0] == 'restaurant':
prev_speaker = ''
prev_utterance = ''
for turn in dialogue['turns']:
count = count + 1
speaker = turn['speaker']
utterance = turn['utterance']
for frame in turn['frames']:
if frame['service'] == 'restaurant':
knowledge = ''
try:
knowledge = '[KNOWLEDGE] '
for slot in frame['slots']:
temp = '%s [EQUAL] %s [SEP] ' % (slot['slot'],slot['value'])
knowledge = knowledge + temp
except:
nothing = 1
try:
if len(knowledge) == 0:
knowledge = '[KNOWLEDGE] '
try:
intent = frame['state']['active_intent']
temp = '%s [EQUAL] %s [SEP] ' % ('active_intent',intent)
knowledge = knowledge + temp
slot_values = frame['state']['slot_values']
for slot in slot_values:
vals = slot_values[slot]
for val in vals:
temp = '%s [EQUAL] %s [SEP] ' % (slot,val)
knowledge = knowledge + temp
except:
nothing = 1
except:
noting = 1
if len(prev_speaker) > 0:
if not bKnowledge:
knowledge = ''
if dataset == 0:
text = '[%s] %s %s [%s] %s [END]' % (prev_speaker,
prev_utterance,
knowledge,
speaker,
utterance)
else:
text = '[%s] %s %s [%s] | %s [END]' % (prev_speaker,
prev_utterance,
knowledge,
speaker,
utterance)
fout.write('%s\n' % (text))
print(text)
prev_speaker = speaker
prev_utterance = utterance
counts.append(count)
count = 0
print(counts)
def main():
make_woz_datasets(True)
make_woz_datasets(False)
gen_mode = 0
gen_labels = ['logits','greedy','beam','top-p']
tuned_model = 2
if tuned_model == 0:
tuned = 'gpt2'
test_name = 'woz.test_a.txt'
elif tuned_model = 1:
tuned = '/home/ddemeter/CS-497/b'
test_name = 'woz.test_b.txt'
else:
tuned = '/home/ddemeter/CS-497/c'
test_name = 'woz.test_c.txt'
tokenizer = GPT2Tokenizer.from_pretrained(tuned)
model = GPT2LMHeadModel.from_pretrained(tuned,pad_token_id=tokenizer.eos_token_id)
model = model.cuda()
model_parameters = filter(lambda p: p.requires_grad, model.parameters())
params = sum([np.prod(p.size()) for p in model_parameters])
print('total parameters = ',params)
metric = load_metric("bleu")
predicts = []
refs = []
best = []
bleus = []
max_len = 0
total = 0
with open(test_name,'rt') as f:
for line in f:
line = line.replace('\n','')
text = line.split('|')
prompt = text[0].strip(' ')
in_ids = tokenizer.encode(prompt,add_special_tokens=True)
if len(in_ids) > max_len:
max_len = len(in_ids)
total = total + 1
max_len = max_len + 32
print('max_len: %d total: %d' % (max_len,total))
obs = 0
with open(test_name,'rt') as f:
for line in f:
line = line.replace('\n','')
text = line.split('|')
prompt = text[0].strip(' ')
ref = text[1].strip(' ')
obs = obs + 1
in_ids = tokenizer.encode(prompt,add_special_tokens=True)
if gen_mode == 0:
seq_len = 0
bDone = False
while not bDone:
input_ids = torch.tensor(in_ids).unsqueeze(0)
input_ids = input_ids.cuda()
outputs = model(input_ids, labels=input_ids)
decoded = []
for i in range(outputs[1].size(1)):
decoded.append(torch.argmax(outputs[1][0][i][:]).item())
decoded = torch.tensor(decoded)
decoded = decoded.cuda()
in_ids.append(decoded[decoded.size(0)-1].item())
input_ids = torch.tensor(in_ids).unsqueeze(0)
text = tokenizer.decode(input_ids[0], skip_special_tokens=True)
tokens = text.split(' ')
if tokens[len(tokens)-1] == '[END]':
bDone = True
if len(tokens) >= max_len:
bDone = True
if gen_mode == 1:
input_ids = torch.tensor(in_ids).unsqueeze(0)
input_ids = input_ids.cuda()
greedy = model.generate(input_ids, max_length=max_len)
text2 = tokenizer.decode(greedy[0], skip_special_tokens=False)
tokens = text2.split()
if gen_mode == 2:
input_ids = torch.tensor(in_ids).unsqueeze(0)
input_ids = input_ids.cuda()
beam = model.generate(input_ids, max_length=max_len, num_beams = 5, early_stopping = True)
text2 = tokenizer.decode(beam[0], skip_special_tokens=False)
tokens = text2.split()
if gen_mode == 3:
input_ids = torch.tensor(in_ids).unsqueeze(0)
input_ids = input_ids.cuda()
top_p = model.generate(input_ids, max_length=max_len, do_sample = True, top_p = 0.90, top_k = 0)
text2 = tokenizer.decode(top_p[0], skip_special_tokens=False)
tokens = text2.split()
first = len(prompt.split(' '))
try:
pos_end = tokens.index('[END]')
except:
pos_end = len(tokens)
try:
pos_enduser = tokens.index('[END][USER]')
except:
pos_enduser = len(tokens)
try:
pos_endsystem = tokens.index('[END][SYSTEM]')
except:
pos_endsystem = len(tokens)
last = min(pos_end,pos_enduser,pos_endsystem,len(tokens))
predict = ' '.join(tokens[first:last])
predictions = [predict.split()]
references = [[ref.split()]]
predicts.append(predictions)
refs.append(references)
try:
results = metric.compute(predictions = predictions, references = references)
bleus.append(results['bleu'])
except:
print('ref: ',ref)
print('pred: ',predict)
if results['bleu'] > 0.01:
print('ref: ',ref)
print('pred: ',predict)
print('BLEU[%d]: %7.3f' % (obs,results['bleu']))
print(' ')
best.append(results['bleu'])
print(best)
if len(best) > 0:
print('avg[%d]: %7.5f' % (len(best),sum(best)/float(len(best))))
print(' ')
results = metric.compute(predictions = predicts, references = refs)
print('Final %s on %s BLEU: %7.3f % 7.3f' % (gen_labels[gen_mode],test_name,results['bleu'],sum(bleus)/511.0))
print(len(predicts),len(refs))
print(' ')
if __name__ == "__main__":
main()