forked from sarthakharne/CASE
-
Notifications
You must be signed in to change notification settings - Fork 0
/
BERTFineTuning.py
224 lines (176 loc) · 6.88 KB
/
BERTFineTuning.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
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
import pandas as pd
import numpy as np
import torch
from torch.utils.data import Dataset, DataLoader, RandomSampler, SequentialSampler
from transformers import BertTokenizer, BertForSequenceClassification, BertForMaskedLM, AutoModel, AutoTokenizer
from transformers import get_scheduler
from transformers import AutoTokenizer, AutoModelForSequenceClassification
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score, f1_score
from tqdm import trange
import random
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import AdamW
import contractions
import re
import time
import gc
from itertools import filterfalse
from tqdm import trange
from tqdm.auto import tqdm
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# SET data_path TO FINETUNING DATASET
data_path = "PATH TO DATA"
# SET base_dir TO PRE TRAINED MODEL PATH
base_dir = 'PATH TO PRE TRAINED MODEL'
# SET you HF TOKEN TO PUSH MODEL TO HUB
hf_token = "ENTER YOUR HUGGINGFACE TOKEN"
def expand_contractions(sentence):
contractions_expanded = [contractions.fix(word) for word in sentence.split()]
return ' '.join(contractions_expanded)
def lower_case(sentence):
return ' '.join([word.lower() for word in sentence.split()])
def remove_punctuation(sentence):
return ' '.join([re.sub(r'[^\w\s]', '', word) for word in sentence.split()])
def preprocess(lst, process=True, min_words=20):
lst[:] = filterfalse(lambda x: len(x.split()) <= min_words, lst)
if process == True:
for i, sent in enumerate(lst):
lst[i] = lower_case(remove_punctuation(expand_contractions(sent)))
return lst
class ClassificationDataset(Dataset):
def __init__(self,
target_column,
tokenizer,
data,
max_len=512):
self.source_column = 'post'
self.target_column = target_column
self.data = data
self.max_len = max_len
self.tokenizer = tokenizer
self.inputs = []
self.targets = []
self._build()
def __len__(self):
return len(self.inputs)
def __getitem__(self, index):
source_ids = self.inputs[index]["input_ids"].squeeze()
src_mask = self.inputs[index]["attention_mask"].squeeze() # might need to squeeze
target = self.targets[index]
return {"source_ids": source_ids, "source_mask": src_mask, "target": target}
def _build(self):
for idx in range(len(self.data)):
input_, target = self.data.loc[idx, self.source_column], self.data.loc[idx, self.target_column]
tokenized_inputs = self.tokenizer.batch_encode_plus(
[input_], max_length=self.max_len, pad_to_max_length=True, truncation=True, return_tensors="pt"
)
self.inputs.append(tokenized_inputs)
if target == True:
self.targets.append(1)
else:
self.targets.append(0)
df_class = pd.read_csv(data_path)
def set_seed(seed):
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
if torch.cuda.is_available():
torch.cuda.manual_seed_all(seed)
set_seed(42)
def get_classification_dataset(target_col, tokenizer, data):
return ClassificationDataset(target_col, tokenizer = tokenizer, data = data)
metric_dict = {}
def process_and_evaluate(
model_loc: str,
tokenizer_loc: str,
column: str,
sanity_check: bool = False
):
# preprocessing
train_sentences = preprocess(list(df_class['questionFull']))
labels = list(df_class[column])
# creating a new dataframe
train_df = pd.DataFrame([])
train_df['post'] = train_sentences
train_df['Disorder'] = labels
del train_sentences
gc.collect()
# getting the model
tokenizer = AutoTokenizer.from_pretrained(tokenizer_loc, token=hf_token)
model = AutoModelForSequenceClassification.from_pretrained(model_loc, token=hf_token).to(device)
# Putting dummy y's for y_train y_test
y = np.ones(train_df.shape[0])
X_train, X_test, y_train, y_test = train_test_split(train_df, y, test_size=0.20, random_state=42)
X_train = X_train.reset_index().drop(columns=["index"])
X_test = X_test.reset_index().drop(columns=["index"])
train_dataset = get_classification_dataset('Disorder', tokenizer, data=X_train)
test_dataset = get_classification_dataset('Disorder', tokenizer, data=X_test)
train_dataloader = DataLoader(train_dataset, batch_size = 32)
test_dataloader = DataLoader(test_dataset, batch_size = 32)
del X_train, X_test, y_train, y_test
gc.collect()
optimizer = AdamW(model.parameters(), lr=5e-5)
num_epochs = 3
num_training_steps = num_epochs * len(train_dataloader)
lr_scheduler = get_scheduler(
name="linear", optimizer=optimizer, num_warmup_steps=0, num_training_steps=num_training_steps
)
progress_bar = tqdm(range(num_training_steps))
model.train()
for epoch in range(num_epochs):
for batch in train_dataloader:
batch = {k: v.to(device) for k, v in batch.items()}
outputs = model(input_ids=batch['source_ids'],attention_mask=batch['source_mask'],labels=batch['target'])
loss = outputs.loss
loss.backward()
optimizer.step()
lr_scheduler.step()
optimizer.zero_grad()
progress_bar.update(1)
if sanity_check == True:
break
if sanity_check == True:
break
model.eval()
y_true = []
y_pred = []
debug_mode = False
with torch.no_grad():
for batch in test_dataloader:
source_ids = batch['source_ids'].to(device)
source_mask = batch['source_mask'].to(device)
targets = batch['target'].to(device)
outputs = model(source_ids, attention_mask=source_mask)
_, predicted = torch.max(outputs.logits, 1)
y_true.extend(targets.cpu().numpy())
if debug_mode:
y_pred.extend([1] * len(predicted))
else:
y_pred.extend(predicted.cpu().numpy())
accuracy = accuracy_score(y_true, y_pred)
f1 = f1_score(y_true, y_pred, average='binary')
print(f'For class: {column} f1 score: {f1} Accuracy: {accuracy}')
arr.append({
"model_name": model_loc,
"class": column,
"f1 score": f1,
"accuracy": accuracy,
})
# COLUMNS TO PROCESS AND EVALUATE
good_cols = ['Anxiety', 'Depression']
arr = []
tokenizer_dir = 'bert-base-uncased'
print('-'*100)
for col in good_cols:
print(f'{col} Started')
torch.cuda.empty_cache()
gc.collect()
try:
process_and_evaluate(base_dir,tokenizer_dir, col)
except Exception as e:
print(f'EXCEPTION OCCURED AT CLASS {col}, for MODEL {base_dir}')
print(e)
continue
print('-'*100)