-
Notifications
You must be signed in to change notification settings - Fork 1
/
get_nli_score.py
161 lines (128 loc) · 7.12 KB
/
get_nli_score.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
import argparse
import numpy as np
import os
import pdb
import gc
import subprocess
from logger import logger
from tqdm import tqdm
from itertools import product
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
from politifactDataloader import Dataset
def run(args):
if not os.path.exists(args.data_root):
logger.error("File not found")
exit()
if not os.path.exists(args.output):
logger.warning("Output dir missing. Creating one...")
os.makedirs(args.output)
torch_device = torch.device('cuda')
dataset = Dataset(args.data_root, concat_before=0, concat_after=0)
verified_claims = dataset.verified_claims
verified_claims = verified_claims.replace(np.nan, '', regex=True)
verified_claims = verified_claims.vclaim
transcripts = dataset.transcripts
transcripts_names = [name for name in transcripts.keys()]
transcripts_names.sort(key=(lambda name: len(transcripts[name])))
max_length = 256
hg_model_hub_name = "ynie/roberta-large-snli_mnli_fever_anli_R1_R2_R3-nli"
# hg_model_hub_name = "ynie/albert-xxlarge-v2-snli_mnli_fever_anli_R1_R2_R3-nli"
# hg_model_hub_name = "ynie/bart-large-snli_mnli_fever_anli_R1_R2_R3-nli"
# hg_model_hub_name = "ynie/electra-large-discriminator-snli_mnli_fever_anli_R1_R2_R3-nli"
# hg_model_hub_name = "ynie/xlnet-large-cased-snli_mnli_fever_anli_R1_R2_R3-nli"
tokenizer = AutoTokenizer.from_pretrained(hg_model_hub_name)
model = AutoModelForSequenceClassification.from_pretrained(hg_model_hub_name)
model = model.to(torch_device)
vclaim_scores_dim = args.elasticsearch_max_length
logger.info(f'Processing debates {args.transcript_start} to {args.transcript_end}')
for transcript_name in tqdm(transcripts_names[args.transcript_start:args.transcript_end]):
transcript = transcripts[transcript_name]
transcript = transcript.replace(np.nan, '', regex=True)
transcript_opath = os.path.join(args.output, '%s.npz'%(transcript_name))
elasticsearch_transcript = np.load(os.path.join(args.elasticsearch_dir, transcript_name+'.npz'), allow_pickle=True)
vclaim_selected_idxs = elasticsearch_transcript[args.measure][0][:, :vclaim_scores_dim].astype('int')
transcript_verified_claims = [verified_claims[sentence_claim_indxs].tolist() for sentence_claim_indxs in vclaim_selected_idxs]
transcript_sentences = transcript.sentence.values.tolist()
logger.info(f'Processing debate {transcript_name} sent. count: {len(transcript_sentences)}')
batch_size=args.batch_size
sentence_claims_pairs = list(zip(transcript_sentences, transcript_verified_claims))
sentence_scores = []
print("batch size: {}".format(batch_size))
print("sentence_claims_pairs size: {}".format(len(sentence_claims_pairs)))
for batch_number in tqdm(range(0, len(sentence_claims_pairs), batch_size)):
logger.info(f'Processing batch number: {batch_number}')
if os.path.exists(transcript_opath):
logger.info(f'File for {transcript_name} already exists')
continue
else:
batch_sentence_pairs = []
for (sentence, sentence_claims) in sentence_claims_pairs[batch_number: batch_number+batch_size]:
sentence_pairs = [list(pair) for pair in product(sentence_claims, [sentence])]
batch_sentence_pairs.extend(sentence_pairs)
del sentence_pairs
print('batch_sentence_length', len(batch_sentence_pairs))
encoded_results = tokenizer.batch_encode_plus(batch_sentence_pairs, max_length=max_length, return_token_type_ids=True, truncation=True, padding=True)
del batch_sentence_pairs
input_ids = torch.Tensor(encoded_results['input_ids']).long().to(torch_device)
token_types = torch.Tensor(encoded_results['token_type_ids']).long().to(torch_device)
att_mask = torch.Tensor(encoded_results['attention_mask']).long().to(torch_device)
del encoded_results
outputs = model(input_ids,
attention_mask=att_mask,
token_type_ids=token_types,
labels=None)
# Since there was not enought memory, I'm computing them into batches
# output_list = []
# pair_batch_size = 20
# for encoded_pair_indx in range(0, len(input_ids),pair_batch_size):
# if encoded_pair_indx % 1000 == 0:
# logger.debug(f'Processing sentence {encoded_pair_indx}')
# outputs = model(input_ids[encoded_pair_indx:encoded_pair_indx+pair_batch_size],
# attention_mask=att_mask[encoded_pair_indx:encoded_pair_indx+pair_batch_size],
# token_type_ids=token_types[encoded_pair_indx:encoded_pair_indx+pair_batch_size],
# labels=None)
# output_list.extend(outputs[0].tolist())
# del outputs
del input_ids, token_types, att_mask
# output_list = torch.Tensor(output_list)
# predicted_probability = torch.softmax(output_list, dim=1)
predicted_probability = torch.softmax(outputs[0], dim=1)
predicted_probability = predicted_probability.tolist()
predicted_probability = [predicted_probability[i:i+vclaim_scores_dim] for i in range(0,len(predicted_probability), vclaim_scores_dim)]
# del output_list
sentence_scores.extend(predicted_probability)
del predicted_probability
sentence_scores = clear_invalid_claims_scores(sentence_scores, vclaim_selected_idxs)
np.savez(transcript_opath, scores=sentence_scores, allow_pickle=True)
def clear_invalid_claims_scores(sentence_scores, vclaim_indxs):
transcript_scores = []
for sent_scores, claim_indxs in list(zip(sentence_scores,vclaim_indxs)):
processed_scores = []
for scores, claim_indx in list(zip(sent_scores, claim_indxs)):
if claim_indx != 0:
processed_scores.append(scores)
else:
empty_scores = np.zeros(3).tolist()
processed_scores.append(empty_scores)
transcript_scores.append(processed_scores)
return transcript_scores
def get_claims_from_idxs(claims_idxs, claims_list):
return [claims_list[idx] for idx in get_unique_claim_idxs(claims_idxs)]
def get_unique_claim_idxs(claims_idxs):
claims_idxs, idxs = np.unique(claims_idxs, return_index=True)
claims_idxs = list(zip(claims_idxs, idxs))
claims_idxs.sort(key=lambda elem: elem[1])
return [elem[0] for elem in claims_idxs]
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Get NLI score values')
parser.add_argument('--data-root', '-d', required=True)
parser.add_argument('--output', '-o', required=True)
parser.add_argument('--elasticsearch-dir', '-e', required=True)
parser.add_argument('--elasticsearch-max-length', '-n', required=True, type=int, help='Dimension of the claims list retrieved from elasticsearch')
parser.add_argument('--measure', '-m', default='vclaim')
parser.add_argument('--transcript-start', '-start', default=0, type=int)
parser.add_argument('--transcript-end', '-end', default=70, type=int)
parser.add_argument('--batch-size', '-s', default=100, type=int)
args = parser.parse_args()
run(args)