-
Notifications
You must be signed in to change notification settings - Fork 3
/
seq2seq.py
392 lines (299 loc) · 17.7 KB
/
seq2seq.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
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
import argparse
import datasets
import evaluate
import glob
import matplotlib.pyplot as plt
import nltk
import numpy as np
import os
import pandas as pd
import random
import re
import torch
from datasets import Dataset, load_dataset, concatenate_datasets
from file_io import *
from huggingface_hub import HfFolder
from nltk.tokenize import sent_tokenize
from sklearn.metrics import f1_score, recall_score, precision_score
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, DataCollatorForSeq2Seq
from transformers import Seq2SeqTrainer, Seq2SeqTrainingArguments
nltk.download("punkt")
device = torch.device('cuda:0') if torch.cuda.is_available() else torch.device('cpu')
print('CUDA: ', torch.cuda.is_available())
emotion_list = ['anger', 'brain dysfunction (forget)', 'emptiness', 'hopelessness', 'loneliness', 'sadness', 'suicide intent', 'worthlessness']
def preprocess_function(sample, padding="max_length"):
# add prefix to the input for t5
inputs = [item for item in sample["text"]]
# tokenize inputs
model_inputs = tokenizer(inputs, max_length=max_source_length, padding=padding, truncation=True)
# tokenize labels
sample["label_id"] = [str(x) for x in sample["label_id"]]
labels = tokenizer(text_target=sample["label_id"], max_length=max_target_length, padding=padding, truncation=True)
# If we are padding here, replace all tokenizer.pad_token_id in the labels by -100 when we want to ignore
# padding in the loss.
if padding == "max_length":
labels["input_ids"] = [
[(l if l != tokenizer.pad_token_id else -100) for l in label] for label in labels["input_ids"]
]
model_inputs["labels"] = labels["input_ids"]
return model_inputs
# helper function to postprocess text
def postprocess_text(preds, labels):
preds = [pred.strip() for pred in preds]
labels = [label.strip() for label in labels]
# rougeLSum expects newline after each sentence
preds = ["\n".join(sent_tokenize(pred)) for pred in preds]
labels = ["\n".join(sent_tokenize(label)) for label in labels]
return preds, labels
def compute_metrics(eval_preds):
preds, labels = eval_preds
#if isinstance(preds, tuple): preds = preds[0]
decoded_preds = tokenizer.batch_decode(preds, skip_special_tokens=True, clean_up_tokenization_spaces = False)
# replace -100 in the labels as we can't decode them.
#print('tokenizer.pad_token_id: ', tokenizer.pad_token_id)
labels = np.where(labels != -100, labels, tokenizer.pad_token_id)
decoded_labels = tokenizer.batch_decode(labels, skip_special_tokens=True)
# Some simple post-processing
decoded_preds, decoded_labels = postprocess_text(decoded_preds, decoded_labels)
#predictions = ["hello there", "general kenobi"]
#references = ["hello there", "general kenobi"]
#rouge = rouge.compute(predictions=decoded_preds, references=decoded_labels)
digits = ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9']
#print('decoded_labels: ', decoded_labels[0])
#decoded_labels = [x.replace(' ', '') for x in decoded_labels] # remove spaces
#print('decoded_labels: ', decoded_labels[0])
decoded_labels = [[int(y) if y in digits else 0 for y in x] for x in decoded_labels]
decoded_labels = [[0]*(len(emotion_list)-len(x)) + x for x in decoded_labels] # if generate not enough
decoded_labels = [x[0:len(emotion_list)] for x in decoded_labels]
decoded_labels = [[0 if y == 0 else 1 for y in x] for x in decoded_labels]
#print('decoded_labels: ', decoded_labels[0])
#print('decoded_preds: ', decoded_preds[0])
#decoded_preds = [x.replace(' ', '') for x in decoded_preds] # remove spaces
decoded_preds = [[int(y) if y in digits else 0 for y in x] for x in decoded_preds]
decoded_preds = [[0]*(len(emotion_list)-len(x)) + x for x in decoded_preds] # if generate not enough
decoded_preds = [x[0:len(emotion_list)] for x in decoded_preds]
decoded_preds = [[0 if y == 0 else 1 for y in x] for x in decoded_preds]
#print('decoded_preds: ', decoded_preds[0])
f1_mi = f1_score(y_true=decoded_labels, y_pred=decoded_preds, average='micro')
re_mi = recall_score(y_true=decoded_labels, y_pred=decoded_preds, average='micro')
pre_mi = precision_score(y_true=decoded_labels, y_pred=decoded_preds, average='micro')
f1_mac = f1_score(y_true=decoded_labels, y_pred=decoded_preds, average='macro')
re_mac = recall_score(y_true=decoded_labels, y_pred=decoded_preds, average='macro')
pre_mac = precision_score(y_true=decoded_labels, y_pred=decoded_preds, average='macro')
result = {}
result['f1_micro'] = f1_mi
result['recall_micro'] = re_mi
result['precision_micro'] = pre_mi
result['f1_macro'] = f1_mac
result['recall_macro'] = re_mac
result['precision_macro'] = pre_mac
prediction_lens = [np.count_nonzero(pred != tokenizer.pad_token_id) for pred in preds]
result["gen_len"] = np.mean(prediction_lens)
return result
def train(train_set, val_set, test_set, tokenizer, model, model_name = 'facebook/bart-base', max_source_length = 64, max_target_length = 8, epochs = 40, batch_size = 4):
# load dataset
'''train_set = datasets.load_dataset('json', data_files = 'dataset/train.json', split="train")
test_set = datasets.load_dataset('json', data_files = 'dataset/test.json', split="train")
val_set = datasets.load_dataset('json', data_files = 'dataset/val.json', split="train")
print(f"Train dataset size: {len(train_set)}")
print(f"Test dataset size: {len(test_set)}")
print(f"Val dataset size: {len(val_set)}")'''
# Load tokenizer
#tokenizer = AutoTokenizer.from_pretrained(model_name)
#dataset['train'] = dataset['train'].shuffle(seed=42).select(range(2000))
#dataset['test'] = dataset['test'].shuffle(seed=42).select(range(1000))
#dataset['train'] = dataset['train'].shuffle(seed=42)
train_df = pd.DataFrame(train_set)
val_df = pd.DataFrame(val_set)
test_df = pd.DataFrame(test_set)
# The maximum total input sequence length after tokenization.
# Sequences longer than this will be truncated, sequences shorter will be padded.
tokenized_inputs = concatenate_datasets([train_set, test_set]).map(lambda x: tokenizer(x["text"], truncation=True), batched=True, remove_columns=['id', 'title', 'post', 'upvotes', 'emotions', 'date', 'text', 'label_id'])
print(f"Max source length: {max_source_length}")
# The maximum total sequence length for target text after tokenization.
# Sequences longer than this will be truncated, sequences shorter will be padded."
tokenized_targets = concatenate_datasets([train_set, test_set]).map(lambda x: tokenizer(str(x["label_id"]), truncation=True), batched=True, remove_columns=['id', 'title', 'post', 'upvotes', 'emotions', 'date', 'text', 'label_id'])
#max_target_length = max([len(str(x)) for x in tokenized_targets["input_ids"]])
print('Tokenized targets: ', tokenized_targets)
print(f"Max target length: {max_target_length}")
tokenized_train_dataset = train_set.map(preprocess_function, batched=True, remove_columns=['id', 'title', 'post', 'upvotes', 'emotions', 'date', 'text', 'label_id'])
print(f"Keys of tokenized dataset: {list(tokenized_train_dataset.features)}")
tokenized_val_dataset = val_set.map(preprocess_function, batched=True, remove_columns=['id', 'title', 'post', 'upvotes', 'emotions', 'date', 'text', 'label_id'])
print(f"Keys of tokenized dataset: {list(tokenized_val_dataset.features)}")
# load model from the hub
#model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
#model = model.to(device)
# Metric
metric_f1 = evaluate.load("f1")
metric_pre = evaluate.load("precision")
metric_re = evaluate.load("recall")
metric_rouge = evaluate.load("rouge")
# we want to ignore tokenizer pad token in the loss
label_pad_token_id = -100
# data collator
data_collator = DataCollatorForSeq2Seq(
tokenizer,
model=model,
label_pad_token_id=label_pad_token_id,
pad_to_multiple_of=8
)
# Hugging Face repository id
repository_id = ''
try:
repository_id = f"{model_name.split('/')[1]}"
except:
repository_id = f"{model_name}"
# define training args
training_args = Seq2SeqTrainingArguments(
gradient_accumulation_steps = 4,
#gradient_checkpointing=True,
output_dir=repository_id,
per_device_train_batch_size=batch_size,
per_device_eval_batch_size=batch_size,
predict_with_generate=True,
fp16=True, # overflows with fp16, for T5 models please hide this parameter
#learning_rate=3e-4,
num_train_epochs=epochs,
# logging & evaluation strategies
logging_dir=f"{repository_id}/logs",
logging_strategy="epoch",
# logging_steps=1000,
evaluation_strategy="epoch",
save_strategy="epoch",
save_total_limit=1,
load_best_model_at_end=False,
metric_for_best_model="eval_f1_macro",
# push to hub parameters
report_to="tensorboard",
generation_max_length = max_target_length,
#push_to_hub=True,
#hub_strategy="every_save",
#hub_model_name=repository_id,
#hub_token=HfFolder.get_token(),
)
# create Trainer instance
trainer = Seq2SeqTrainer(
model=model,
args=training_args,
data_collator=data_collator,
train_dataset=tokenized_train_dataset,
eval_dataset=tokenized_val_dataset,
compute_metrics=compute_metrics,
)
trainer.train()
#trainer.evaluate()
def test(dataset, model_name, model, tokenizer, input_file = 'dataset/test.json', \
batch_size = 4, max_len = 128, min_len = 1):
if (len(dataset) == 0): # load dataset if not given
dataset = read_list_from_jsonl_file(input_file)
pred_list = []
for i in range(0, len(dataset), batch_size):
n_batch = 0
if (len(dataset)%batch_size != 0): n_batch = len(dataset)//batch_size + 1
else: n_batch = len(dataset)//batch_size
sys.stdout.write('Infer batch: %d/%d \t Model: %s \r' % (i//batch_size + 1, n_batch, model_name))
#sys.stdout.flush()
subset = dataset[i:i + batch_size]
texts = [item['text'] for item in subset]
#print(texts)
inputs = tokenizer(texts, padding = "max_length", truncation = True, max_length = max_len, \
return_tensors = 'pt').to(device)
outputs = []
with torch.no_grad():
outputs = model.generate(**inputs, max_length = max_len, min_length = min_len, \
num_beams = 4, do_sample = False, return_dict_in_generate = True, output_scores = True)
preds = tokenizer.batch_decode(outputs.sequences, skip_special_tokens = True)
preds = [[x] for x in preds]
pred_list += preds
pred_list = [[x for x in pred][0].strip() for pred in pred_list] # use strip() to remove spaces
label_list = [str(item['label_id']) for item in dataset]
digits = ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9']
# ['hola' if i == 1 else '' for i in list_num]
label_list = [[int(y) if y in digits else 0 for y in x] for x in label_list]
label_list = [[0]*(len(emotion_list)-len(x)) + x for x in label_list] # if generate not enough
label_list = [x[0:len(emotion_list)] for x in label_list]
label_list = [[0 if y == 0 else 1 for y in x] for x in label_list]
print('label_list: ', label_list)
pred_list = [[int(y) if y in digits else 0 for y in x] for x in pred_list]
pred_list = [[0]*(len(emotion_list)-len(x)) + x for x in pred_list] # if generate not enough
pred_list = [x[0:len(emotion_list)] for x in pred_list]
pred_list = [[0 if y == 0 else 1 for y in x] for x in pred_list]
print('pred_list: ', pred_list)
f1_mi = f1_score(y_true=label_list, y_pred=pred_list, average='micro')
re_mi = recall_score(y_true=label_list, y_pred=pred_list, average='micro')
pre_mi = precision_score(y_true=label_list, y_pred=pred_list, average='micro')
f1_mac = f1_score(y_true=label_list, y_pred=pred_list, average='macro')
re_mac = recall_score(y_true=label_list, y_pred=pred_list, average='macro')
pre_mac = precision_score(y_true=label_list, y_pred=pred_list, average='macro')
result = {}
result['f1_micro'] = f1_mi
result['recall_micro'] = re_mi
result['precision_micro'] = pre_mi
result['f1_macro'] = f1_mac
result['recall_macro'] = re_mac
result['precision_macro'] = pre_mac
print('Emotion: All')
print(str(round(f1_mac, 2)) + ' & ' + str(round(pre_mac, 2)) + ' & ' + str(round(re_mac, 2)) + ' & ' + str(round(f1_mi, 2)) + ' & ' + str(round(pre_mi, 2)) + ' & ' + str(round(re_mi, 2)))
print('----------------------------------------')
for index, emotion in enumerate(emotion_list):
print('emotion: ', emotion)
temp_label_list = [x[index] for x in label_list]
temp_pred_list = [x[index] for x in pred_list]
f1_mi = f1_score(y_true=temp_label_list, y_pred=temp_pred_list, average='micro')
re_mi = recall_score(y_true=temp_label_list, y_pred=temp_pred_list, average='micro')
pre_mi = precision_score(y_true=temp_label_list, y_pred=temp_pred_list, average='micro')
f1_mac = f1_score(y_true=temp_label_list, y_pred=temp_pred_list, average='macro')
re_mac = recall_score(y_true=temp_label_list, y_pred=temp_pred_list, average='macro')
pre_mac = precision_score(y_true=temp_label_list, y_pred=temp_pred_list, average='macro')
result['f1_micro_' + emotion] = f1_mi
result['recall_micro_' + emotion] = re_mi
result['precision_micro_' + emotion] = pre_mi
result['f1_macro_' + emotion] = f1_mac
result['recall_macro_' + emotion] = re_mac
result['precision_macro_' + emotion] = pre_mac
print(str(round(f1_mac, 2)) + ' & ' + str(round(pre_mac, 2)) + ' & ' + str(round(re_mac, 2)) + ' & ' + str(round(f1_mi, 2)) + ' & ' + str(round(pre_mi, 2)) + ' & ' + str(round(re_mi, 2)))
print('----------------------------------------')
print('result: ', result)
return result
def main(args):
if (args.mode == 'train'):
train_set = datasets.load_dataset('json', data_files = args.train_path, split="train")
test_set = datasets.load_dataset('json', data_files = args.test_path, split="train")
val_set = datasets.load_dataset('json', data_files = args.val_path, split="train")
print(f"Train dataset size: {len(train_set)}")
print(f"Test dataset size: {len(test_set)}")
print(f"Val dataset size: {len(val_set)}")
model = AutoModelForSeq2SeqLM.from_pretrained(args.model_name)
model.to(device)
train(train_set, val_set, test_set, tokenizer, model, model_name = args.model_name, max_source_length = args.max_source_length, max_target_length = args.max_target_length, epochs = args.epochs, batch_size = args.batch_size)
elif (args.mode == 'test'):
model = AutoModelForSeq2SeqLM.from_pretrained(args.model_path)
model.to(device)
model.eval()
test([], args.model_name, model, tokenizer, input_file = args.test_path, batch_size = args.test_batch_size, max_len = args.max_source_length, min_len = args.min_target_length)
#...............................................................................
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Training Parameter')
parser.add_argument('--mode', type=str, default='train') # or test
parser.add_argument('--model_name', type=str, default='facebook/bart-base') # or test
parser.add_argument('--train_path', type=str, default='dataset/train.json')
parser.add_argument('--test_path', type=str, default='dataset/test.json')
parser.add_argument('--val_path', type=str, default='dataset/val.json')
parser.add_argument('--epochs', type=int, default=40)
parser.add_argument('--batch_size', type=int, default=4)
parser.add_argument('--test_batch_size', type=int, default=16)
parser.add_argument('--max_source_length', type=int, default=256)
parser.add_argument('--max_target_length', type=int, default=8)
parser.add_argument('--min_target_length', type=int, default=1)
parser.add_argument('--model_path', type=str, default='bart-base\checkpoint-452')
parser.add_argument('--test_file', type=str, default='dataset/test.json')
args = parser.parse_args()
global tokenizer
tokenizer = AutoTokenizer.from_pretrained(args.model_name, do_lower_case = False, add_prefix_space = True)
global max_source_length
max_source_length = args.max_source_length
global max_target_length
max_target_length = args.max_target_length
main(args)
# python seq2seq.py --mode "train" --model_name "facebook/bart-base" --train_path "dataset/train.json" --val_path "dataset/val.json" --test_path "dataset/test.json" --epochs 25 --batch_size 4 --max_source_length 256
# python seq2seq.py --mode "test" --model_name "facebook/bart-base" --model_path "bart-base\checkpoint-1321" --test_path "dataset/test.json" --test_batch_size 4 --max_source_length 256 --min_target_length 1