File size: 9,688 Bytes
2359bda
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
The script shows how to train Augmented SBERT (In-Domain) strategy for STSb dataset with Semantic Search Sampling.


Methodology:
Three steps are followed for AugSBERT data-augmentation strategy with Semantic Search - 
    1. Fine-tune cross-encoder (BERT) on gold STSb dataset
    2. Fine-tuned Cross-encoder is used to label on Sem. Search sampled unlabeled pairs (silver STSb dataset) 
    3. Bi-encoder (SBERT) is finally fine-tuned on both gold + silver STSb dataset

Citation: https://arxiv.org/abs/2010.08240

Usage:
python train_sts_indomain_semantic.py

OR
python train_sts_indomain_semantic.py pretrained_transformer_model_name top_k

python train_sts_indomain_semantic.py bert-base-uncased 3
"""
from torch.utils.data import DataLoader
from sentence_transformers import models, losses, util
from sentence_transformers import LoggingHandler, SentenceTransformer
from sentence_transformers.cross_encoder import CrossEncoder
from sentence_transformers.cross_encoder.evaluation import CECorrelationEvaluator
from sentence_transformers.evaluation import EmbeddingSimilarityEvaluator
from sentence_transformers.readers import InputExample
from datetime import datetime
import logging
import csv
import torch
import tqdm
import sys
import math
import gzip
import os

#### Just some code to print debug information to stdout
logging.basicConfig(format='%(asctime)s - %(message)s',
                    datefmt='%Y-%m-%d %H:%M:%S',
                    level=logging.INFO,
                    handlers=[LoggingHandler()])
#### /print debug information to stdout


#You can specify any huggingface/transformers pre-trained model here, for example, bert-base-uncased, roberta-base, xlm-roberta-base
model_name = sys.argv[1] if len(sys.argv) > 1 else 'bert-base-uncased'
top_k = int(sys.argv[2]) if len(sys.argv) > 2 else 3

batch_size = 16
num_epochs = 1
max_seq_length = 128

###### Read Datasets ######

#Check if dataset exsist. If not, download and extract  it
sts_dataset_path = 'datasets/stsbenchmark.tsv.gz'

if not os.path.exists(sts_dataset_path):
    util.http_get('https://sbert.net/datasets/stsbenchmark.tsv.gz', sts_dataset_path)

cross_encoder_path = 'output/cross-encoder/stsb_indomain_'+model_name.replace("/", "-")+'-'+datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
bi_encoder_path = 'output/bi-encoder/stsb_augsbert_SS_'+model_name.replace("/", "-")+'-'+datetime.now().strftime("%Y-%m-%d_%H-%M-%S")

###### Cross-encoder (simpletransformers) ######
logging.info("Loading cross-encoder model: {}".format(model_name))
# Use Huggingface/transformers model (like BERT, RoBERTa, XLNet, XLM-R) for cross-encoder model
cross_encoder = CrossEncoder(model_name, num_labels=1)


###### Bi-encoder (sentence-transformers) ######
logging.info("Loading bi-encoder model: {}".format(model_name))
# Use Huggingface/transformers model (like BERT, RoBERTa, XLNet, XLM-R) for mapping tokens to embeddings
word_embedding_model = models.Transformer(model_name, max_seq_length=max_seq_length)

# Apply mean pooling to get one fixed sized sentence vector
pooling_model = models.Pooling(word_embedding_model.get_word_embedding_dimension(),
                               pooling_mode_mean_tokens=True,
                               pooling_mode_cls_token=False,
                               pooling_mode_max_tokens=False)

bi_encoder = SentenceTransformer(modules=[word_embedding_model, pooling_model])


#####################################################
#
# Step 1: Train cross-encoder model with STSbenchmark
#
#####################################################

logging.info("Step 1: Train cross-encoder: {} with STSbenchmark (gold dataset)".format(model_name))

gold_samples = []
dev_samples = []
test_samples = []

with gzip.open(sts_dataset_path, 'rt', encoding='utf8') as fIn:
    reader = csv.DictReader(fIn, delimiter='\t', quoting=csv.QUOTE_NONE)
    for row in reader:
        score = float(row['score']) / 5.0  # Normalize score to range 0 ... 1

        if row['split'] == 'dev':
            dev_samples.append(InputExample(texts=[row['sentence1'], row['sentence2']], label=score))
        elif row['split'] == 'test':
            test_samples.append(InputExample(texts=[row['sentence1'], row['sentence2']], label=score))
        else:
            #As we want to get symmetric scores, i.e. CrossEncoder(A,B) = CrossEncoder(B,A), we pass both combinations to the train set
            gold_samples.append(InputExample(texts=[row['sentence1'], row['sentence2']], label=score))
            gold_samples.append(InputExample(texts=[row['sentence2'], row['sentence1']], label=score))


# We wrap gold_samples (which is a List[InputExample]) into a pytorch DataLoader
train_dataloader = DataLoader(gold_samples, shuffle=True, batch_size=batch_size)


# We add an evaluator, which evaluates the performance during training
evaluator = CECorrelationEvaluator.from_input_examples(dev_samples, name='sts-dev')

# Configure the training
warmup_steps = math.ceil(len(train_dataloader) * num_epochs * 0.1) #10% of train data for warm-up
logging.info("Warmup-steps: {}".format(warmup_steps))

# Train the cross-encoder model
cross_encoder.fit(train_dataloader=train_dataloader,
          evaluator=evaluator,
          epochs=num_epochs,
          evaluation_steps=1000,
          warmup_steps=warmup_steps,
          output_path=cross_encoder_path)

############################################################################
#
# Step 2: Find silver pairs to label
#
############################################################################

#### Top k similar sentences to be retrieved ####
#### Larger the k, bigger the silver dataset ####

logging.info("Step 2.1: Generate STSbenchmark (silver dataset) using pretrained SBERT \
    model and top-{} semantic search combinations".format(top_k))

silver_data = []
sentences = set()

for sample in gold_samples:
    sentences.update(sample.texts)

sentences = list(sentences) # unique sentences
sent2idx = {sentence: idx for idx, sentence in enumerate(sentences)} # storing id and sentence in dictionary
duplicates = set((sent2idx[data.texts[0]], sent2idx[data.texts[1]]) for data in gold_samples) # not to include gold pairs of sentences again


# For simplicity we use a pretrained model
semantic_model_name = 'paraphrase-MiniLM-L6-v2'
semantic_search_model = SentenceTransformer(semantic_model_name)
logging.info("Encoding unique sentences with semantic search model: {}".format(semantic_model_name))

# encoding all unique sentences present in the training dataset
embeddings = semantic_search_model.encode(sentences, batch_size=batch_size, convert_to_tensor=True)

logging.info("Retrieve top-{} with semantic search model: {}".format(top_k, semantic_model_name))

# retrieving top-k sentences given a sentence from the dataset
progress = tqdm.tqdm(unit="docs", total=len(sent2idx))
for idx in range(len(sentences)):
    sentence_embedding = embeddings[idx]
    cos_scores = util.cos_sim(sentence_embedding, embeddings)[0]
    cos_scores = cos_scores.cpu()
    progress.update(1)

    #We use torch.topk to find the highest 5 scores
    top_results = torch.topk(cos_scores, k=top_k+1)
    
    for score, iid in zip(top_results[0], top_results[1]):
        if iid != idx and (iid, idx) not in duplicates:
            silver_data.append((sentences[idx], sentences[iid]))
            duplicates.add((idx,iid))

progress.reset()
progress.close()

logging.info("Length of silver_dataset generated: {}".format(len(silver_data)))
logging.info("Step 2.2: Label STSbenchmark (silver dataset) with cross-encoder: {}".format(model_name))
cross_encoder = CrossEncoder(cross_encoder_path)
silver_scores = cross_encoder.predict(silver_data)

# All model predictions should be between [0,1]
assert all(0.0 <= score <= 1.0 for score in silver_scores)

############################################################################################
#
# Step 3: Train bi-encoder model with both STSbenchmark and labeled AllNlI - Augmented SBERT
#
############################################################################################

logging.info("Step 3: Train bi-encoder: {} with STSbenchmark (gold + silver dataset)".format(model_name))

# Convert the dataset to a DataLoader ready for training
logging.info("Read STSbenchmark gold and silver train dataset")
silver_samples = list(InputExample(texts=[data[0], data[1]], label=score) for \
    data, score in zip(silver_data, silver_scores))


train_dataloader = DataLoader(gold_samples + silver_samples, shuffle=True, batch_size=batch_size)
train_loss = losses.CosineSimilarityLoss(model=bi_encoder)

logging.info("Read STSbenchmark dev dataset")
evaluator = EmbeddingSimilarityEvaluator.from_input_examples(dev_samples, name='sts-dev')

# Configure the training.
warmup_steps = math.ceil(len(train_dataloader) * num_epochs * 0.1) #10% of train data for warm-up
logging.info("Warmup-steps: {}".format(warmup_steps))

# Train the bi-encoder model
bi_encoder.fit(train_objectives=[(train_dataloader, train_loss)],
          evaluator=evaluator,
          epochs=num_epochs,
          evaluation_steps=1000,
          warmup_steps=warmup_steps,
          output_path=bi_encoder_path
          )

#################################################################################
#
# Evaluate cross-encoder and Augmented SBERT performance on STS benchmark dataset
#
#################################################################################

# load the stored augmented-sbert model
bi_encoder = SentenceTransformer(bi_encoder_path)
test_evaluator = EmbeddingSimilarityEvaluator.from_input_examples(test_samples, name='sts-test')
test_evaluator(bi_encoder, output_path=bi_encoder_path)