-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathprediction_analysis.py
144 lines (118 loc) · 5.93 KB
/
prediction_analysis.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
# Prediction results analysis last updated: 2023.06
import os
import argparse
from collections import defaultdict
import torch
from torch.utils.data import DataLoader
from transformers import AutoTokenizer, AutoConfig
from utils.file_reader import ner_reader, pos_reader
from utils.metrics import NERMetric, POSMetric
from PipeLine.glue_utils_transformer import SeqDataset, CollateFnSeq
READER = {
'absa': ner_reader,
'NER': ner_reader,
'pos': pos_reader
}
METRIC = {
'absa': NERMetric,
'NER': NERMetric,
'pos': POSMetric
}
def batch_to_device(batch, device):
for key, value in batch.items():
batch[key] = batch[key].to(device=device)
def init_args():
arguments = argparse.ArgumentParser()
arguments.add_argument('--task_type', type=str, default='NER')
arguments.add_argument('--dataset_name', type=str, default='weibo')
arguments.add_argument('--cache_dir', type=str, default='/home/cs.aau.dk/ut65zx/bert-base-chinese')
arguments.add_argument('--model_base', type=str, default='/home/cs.aau.dk/ut65zx')
arguments.add_argument('--dataset_base', type=str, default='Dataset')
arguments.add_argument('--context_model', type=str, default='bert-base-chinese_13.pth')
arguments.add_argument('--tagger_model', type=str, default='bert-base-chinese_11.pth')
arguments.add_argument('--result_dir', type=str, default='result/')
arguments.add_argument('--model_name', type=str, default='bert-base-chinese')
arguments.add_argument('--bert_size', type=int, default=768)
arguments.add_argument('--batch_size', type=int, default=16)
arguments.add_argument('--device', type=str, default='cuda:0')
args = arguments.parse_args()
return args
def judge(context_model, tagger_model, data_loader, metric, result_dir, device, collect_fn):
context_model.eval()
tagger_model.eval()
batch_id = 0
for batch in data_loader:
first = True
batch_to_device(batch, device)
context_output, (global_tensor, local_tensor) = context_model(**batch)
tagger_output, _ = tagger_model(**batch)
context_pred = torch.argmax(context_output[1], dim=-1)
tagger_pred = torch.argmax(tagger_output[1], dim=-1)
gold_label_ids = batch['labels']
gold_label, gold_text = collect_fn.batch_labels, collect_fn.batch_texts
context_pred_ = metric.get_entity_batch(context_pred, gold=gold_label_ids, ignore_index=-100)
tagger_pred_ = metric.get_entity_batch(tagger_pred, gold=gold_label_ids, ignore_index=-100)
gold_label_ = metric.get_entity_batch(gold_label_ids, gold=gold_label_ids, ignore_index=-100)
batch_len = len(gold_text)
for i in range(batch_len):
gold_label_item, text_item = gold_label[i], gold_text[i]
context_item, tagger_item = context_pred_[i], tagger_pred_[i]
if len(context_item) == len(gold_label_item):
if context_item != tagger_item:
if first:
torch.save(global_tensor.cpu(), os.path.join(result_dir, f'global_tensor_{batch_id}.pt'))
torch.save(local_tensor.cpu(), os.path.join(result_dir, f'local_tensor_{batch_id}.pt'))
first = False
difference_file = os.path.join(result_dir, f'difference_{str(batch_id)}.txt')
with open(difference_file, 'a+') as f:
f.write(f'Itme number: {i + 1}')
f.write('\n')
seq_len = len(gold_label_item)
for j in range(seq_len):
f.write(text_item[j] + '\t' + gold_label_item[j] + '\t'
+ context_item[j] + '\t' + tagger_item[j])
f.write('\n')
f.write('\n')
else:
print(i)
print(text_item)
print(context_item)
print(gold_label_item)
batch_id += 1
def main():
args = init_args()
if not os.path.exists(args.result_dir):
os.mkdir(args.result_dir)
context_model_path = os.path.join(args.model_base, 'saved_model-tagger-context', args.context_model)
tagger_model_path = os.path.join(args.model_base, 'saved_model-tagger', args.tagger_model)
test_source = os.path.join(args.dataset_base, args.task_type, args.dataset_name, 'test.txt')
train_source = os.path.join(args.dataset_base, args.task_type, args.dataset_name, 'train.txt')
reader = READER.get(args.task_type, ner_reader)
metric_class = METRIC.get(args.task_type, NERMetric)
device = torch.device(args.device)
if os.path.exists(args.cache_dir):
configer = AutoConfig.from_pretrained(args.cache_dir, hidden_size=args.bert_size)
tokenizer = AutoTokenizer.from_pretrained(args.cache_dir)
else:
configer = AutoConfig.from_pretrained(args.model_name)
tokenizer = AutoTokenizer.from_pretrained(args.model_name)
os.makedirs(args.cache_dir)
configer.save_pretrained(args.cache_dir)
tokenizer.save_pretrained(args.cache_dir)
dataset_ = SeqDataset(test_source, read_method=reader)
train_dataset_ = SeqDataset(train_source, read_method=reader)
label_list = train_dataset_.get_label()
label2idx = defaultdict()
label2idx.default_factory = label2idx.__len__
idx2label = {}
for label_name in label_list:
idx = label2idx[label_name]
idx2label[idx] = label_name
collect_fn = CollateFnSeq(tokenizer=tokenizer, label2idx=label2idx)
loader = DataLoader(dataset_, collate_fn=collect_fn, batch_size=args.batch_size)
context_model, tagger_model = torch.load(context_model_path), torch.load(tagger_model_path)
metric = metric_class(id2token=idx2label)
judge(context_model=context_model, tagger_model=tagger_model, data_loader=loader,
metric=metric, result_dir=args.result_dir, device=device, collect_fn=collect_fn)
if __name__ == '__main__':
main()