-
Notifications
You must be signed in to change notification settings - Fork 17
/
Copy pathtest.py
175 lines (153 loc) · 6 KB
/
test.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
# Author: yxsu
import sys
import torch
import argparse
sys.path.append('..')
from bert import BERTLM
from data import SEP, MASK
from main import myModel
import numpy as np
import time
mstime = lambda: int(round(time.time() * 1000))
def extract_parameters(ckpt_path):
model_ckpt = torch.load(ckpt_path, map_location='cpu')
bert_args = model_ckpt['bert_args']
model_args = model_ckpt['args']
bert_vocab = model_ckpt['bert_vocab']
model_parameters = model_ckpt['model']
return bert_args, model_args, bert_vocab, model_parameters
def init_empty_bert_model(bert_args, bert_vocab, gpu_id):
bert_model = BERTLM(gpu_id, bert_vocab, bert_args.embed_dim, bert_args.ff_embed_dim, bert_args.num_heads, \
bert_args.dropout, bert_args.layers, bert_args.approx)
return bert_model
def init_sequence_tagging_model(empty_bert_model, args, bert_args, gpu_id, bert_vocab, model_parameters):
number_class = args.number_class
embedding_size = bert_args.embed_dim
batch_size = args.batch_size
dropout = args.dropout
device = gpu_id
vocab = bert_vocab
loss_type = args.loss_type
seq_tagging_model = myModel(empty_bert_model, number_class, embedding_size, batch_size, dropout,
device, vocab, loss_type)
seq_tagging_model.load_state_dict(model_parameters)
return seq_tagging_model
def get_tag_mask_matrix(batch_text_list):
tag_matrix = []
mask_matrix = []
batch_size = len(batch_text_list)
max_len = 0
for instance in batch_text_list:
max_len = max(len(instance), max_len)
max_len += 1 # 1 for [CLS]
for i in range(batch_size):
one_text_list = batch_text_list[i]
one_tag = list(np.zeros(max_len).astype(int))
tag_matrix.append(one_tag)
one_mask = [1]
one_valid_len = len(batch_text_list[i])
for j in range(one_valid_len):
one_mask.append(1)
len_diff = max_len - len(one_mask)
for _ in range(len_diff):
one_mask.append(0)
mask_matrix.append(one_mask)
assert len(one_mask) == len(one_tag)
return np.array(tag_matrix), np.array(mask_matrix)
def join_str(in_list):
out_str = ''
for token in in_list:
out_str += str(token) + ''
return out_str.strip()
def predict_one_text_split(text_split_list, seq_tagging_model, label_dict):
# text_split_list is a list of tokens ['word1', 'word2', ...]
text_list = [text_split_list]
tag_matrix, mask_matrix = get_tag_mask_matrix(text_list)
decode_result, _, _, _ = seq_tagging_model(text_list, mask_matrix, tag_matrix, fine_tune = False)
valid_text_len = len(text_split_list)
valid_decode_result = decode_result[0][1: valid_text_len + 1]
tag_result = []
for token in valid_decode_result:
tag_result.append(label_dict[int(token)])
return tag_result
#return valid_decode_result
def get_text_split_list(text, max_len):
result_list = []
text_list = [w for w in text] + [SEP]
valid_len = len(text_list)
split_num = (len(text_list) // max_len) + 1
if split_num == 1:
result_list = [text_list]
else:
b_idx = 0
e_idx = 1
for i in range(max_len):
b_idx = i * max_len
e_idx = (i + 1) * max_len
result_list.append(text_list[b_idx:e_idx])
if e_idx < valid_len:
result_list.append(text_list[e_idx:])
else:
pass
return result_list
def predict_one_text(text, max_len, seq_tagging_model, label_dict):
text_split_list = get_text_split_list(text, max_len)
all_text_result = []
all_decode_result = []
for one_text_list in text_split_list:
one_decode_result = predict_one_text_split(one_text_list, seq_tagging_model, label_dict)
all_text_result.extend(one_text_list)
all_decode_result.extend(one_decode_result)
result_text = join_str(all_text_result)
tag_predict_result = join_str(all_decode_result)
return tag_predict_result
def get_label_dict(label_path):
label_dict = {}
with open(label_path, 'r', encoding = 'utf8') as i:
lines = i.readlines()
for l in lines:
content_list = l.strip('\n').split()
label_id = int(content_list[1])
label = content_list[0]
label_dict[label_id] = label
return label_dict
def parse_config():
parser = argparse.ArgumentParser()
parser.add_argument('--ckpt_path', type=str)
parser.add_argument('--test_data',type=str)
parser.add_argument('--out_path',type=str)
parser.add_argument('--gpu_id',type=int, default=0)
parser.add_argument('--max_len',type=int)
return parser.parse_args()
if __name__ == "__main__":
args = parse_config()
ckpt_path = args.ckpt_path
test_data = args.test_data
out_path = args.out_path
gpu_id = args.gpu_id
max_len = args.max_len
print("loading..")
bert_args, model_args, bert_vocab, model_parameters = extract_parameters(ckpt_path)
label_dict = {}
for lid, label in enumerate(bert_vocab._idx2token):
label_dict[lid] = label
model_args.number_class = len(label_dict)
empty_bert_model = init_empty_bert_model(bert_args, bert_vocab, gpu_id)
seq_tagging_model = init_sequence_tagging_model(empty_bert_model, model_args,
bert_args, gpu_id, bert_vocab, model_parameters)
seq_tagging_model.cuda(gpu_id)
print("eval...")
seq_tagging_model.eval()
with torch.no_grad():
with open(out_path, 'w', encoding = 'utf8') as o:
with open(test_data, 'r', encoding = 'utf8') as i:
start = mstime()
lines = i.readlines()
for l in lines:
content_list = l.strip().split('\t')
text = content_list[0]
gold = content_list[1]
res = predict_one_text(text, max_len, seq_tagging_model, label_dict)
res = res.replace(SEP, '').strip()
o.writelines(text + "\t" + res + "\t" + gold + "\n")
print(mstime()-start)