-
Notifications
You must be signed in to change notification settings - Fork 12
/
Copy pathevaluate.py
152 lines (118 loc) · 6.5 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
from __future__ import print_function
import argparse
import time
import torch
import numpy as np
import dataLoader
import helper
import tempfile
from helper import tokens_to_sentences
from reinforce import return_summary_index
from rougefonc import from_summary_index_generate_hyp_ref, RougeTest_pyrouge, RougeTest_rouge
np.set_printoptions(precision=4, suppress=True)
def reinforce_loss(probs, doc, id=0,
max_num_of_sents=3, max_num_of_bytes=-1,
std_rouge=False, rouge_metric="all", compute_score=True):
# sample sentences
probs_numpy = probs.data.cpu().numpy()
probs_numpy = np.reshape(probs_numpy, len(probs_numpy))
max_num_of_sents = min(len(probs_numpy), max_num_of_sents) # max of sents# in doc and sents# in summary
rl_baseline_summary_index, _ = return_summary_index(probs_numpy, probs,
sample_method="greedy", max_num_of_sents=max_num_of_sents)
rl_baseline_summary_index = sorted(rl_baseline_summary_index)
rl_baseline_hyp, rl_baseline_ref = from_summary_index_generate_hyp_ref(doc, rl_baseline_summary_index)
lead3_hyp, lead3_ref = from_summary_index_generate_hyp_ref(doc, range(max_num_of_sents))
if std_rouge:
rl_baseline_reward = RougeTest_pyrouge(rl_baseline_ref, rl_baseline_hyp, id=id, rouge_metric=rouge_metric,
compute_score=compute_score, path=os.path.join('./result/rl'),
max_num_of_bytes=max_num_of_bytes)
lead3_reward = RougeTest_pyrouge(lead3_ref, lead3_hyp, id=id, rouge_metric=rouge_metric,
compute_score=compute_score, path=os.path.join('./result/lead'),
max_num_of_bytes=max_num_of_bytes)
else:
rl_baseline_reward = RougeTest_rouge(rl_baseline_ref, rl_baseline_hyp, rouge_metric,
max_num_of_bytes=max_num_of_bytes)
lead3_reward = RougeTest_rouge(lead3_ref, lead3_hyp, rouge_metric, max_num_of_bytes=max_num_of_bytes)
return rl_baseline_reward, lead3_reward
def ext_model_eval(model, vocab, args, eval_data="test"):
print("loading data %s" % eval_data)
model.eval()
data_loader = dataLoader.PickleReader(args.data_dir)
eval_rewards, lead3_rewards = [], []
data_iter = data_loader.chunked_data_reader(eval_data)
print("doing model evaluation on %s" % eval_data)
for phase, dataset in enumerate(data_iter):
for step, docs in enumerate(dataLoader.BatchDataLoader(dataset, shuffle=False)):
print("Done %2d chunck, %4d/%4d doc\r" % (phase+1, step + 1, len(dataset)), end='')
doc = docs[0]
doc.content = tokens_to_sentences(doc.content)
doc.summary = tokens_to_sentences(doc.summary)
if len(doc.content) == 0 or len(doc.summary) == 0:
continue
# if doc.content[0].find('CNN') >= 0:
# args.oracle_length = 3
# else:
# args.oracle_length = 4
if args.oracle_length == -1: # use true oracle length
oracle_summary_sent_num = len(doc.summary)
else:
oracle_summary_sent_num = args.oracle_length
x = helper.prepare_data(doc, vocab)
if min(x.shape) == 0:
continue
sents = torch.autograd.Variable(torch.from_numpy(x)).cuda()
outputs = model(sents)
compute_score = (step == len(dataset) - 1) or (args.std_rouge is False)
if eval_data == "test":
# try:
reward, lead3_r = reinforce_loss(outputs, doc, id=phase * 1000 + step,
max_num_of_sents=oracle_summary_sent_num,
max_num_of_bytes=args.length_limit,
std_rouge=args.std_rouge, rouge_metric="all",
compute_score=compute_score)
else:
reward, lead3_r = reinforce_loss(outputs, doc, id=phase * 1000 + step,
max_num_of_sents=oracle_summary_sent_num,
max_num_of_bytes=args.length_limit,
std_rouge=args.std_rouge, rouge_metric=args.rouge_metric,
compute_score=compute_score)
if compute_score:
eval_rewards.append(reward)
lead3_rewards.append(lead3_r)
avg_eval_r = np.mean(eval_rewards, axis=0)
avg_lead3_r = np.mean(lead3_rewards, axis=0)
print('model %s reward in %s:' % (args.rouge_metric, eval_data))
print(avg_eval_r)
print(avg_lead3_r)
return avg_eval_r, avg_lead3_r
if __name__ == '__main__':
from dataLoader import *
torch.manual_seed(233)
parser = argparse.ArgumentParser()
parser.add_argument('--vocab_file', type=str, default='../data/CNN_DM_pickle_data/vocab_100d.p')
parser.add_argument('--data_dir', type=str, default='../data/CNN_DM_pickle_data/')
parser.add_argument('--model_file', type=str, default='../model/summary.ext')
parser.add_argument('--device', type=int, default=0,
help='select GPU')
parser.add_argument('--std_rouge', action='store_true')
parser.add_argument('--oracle_length', type=int, default=3,
help='-1 for giving actual oracle number of sentences'
'otherwise choose a fixed number of sentences')
parser.add_argument('--rouge_metric', type=str, default='all')
parser.add_argument('--rl_baseline_method', type=str, default="greedy",
help='greedy, global_avg,batch_avg,or none')
parser.add_argument('--length_limit', type=int, default=-1,
help='length limit output')
args = parser.parse_args()
torch.cuda.set_device(args.device)
print('generate config')
with open(args.vocab_file, "rb") as f:
vocab = pickle.load(f)
print(vocab)
print("loading existing model%s" % args.model_file)
extract_net = torch.load(args.model_file, map_location=lambda storage, loc: storage)
extract_net.cuda()
print("finish loading and evaluate model %s" % args.model_file)
start_time = time.time()
ext_model_eval(extract_net, vocab, args, eval_data="test")
print('Test time:', time.time() - start_time)