Created
August 19, 2023 11:26
-
-
Save hadifar/de0923a839387689dd914dad602143d3 to your computer and use it in GitHub Desktop.
simple classifier code
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import evaluate | |
import numpy as np | |
import random | |
import torch | |
from datasets import load_dataset | |
from transformers import AutoTokenizer | |
from transformers import DataCollatorWithPadding | |
from transformers import AutoModelForSequenceClassification, TrainingArguments, Trainer | |
CACHE_DIR = 'cache/' | |
ROOT_DIR = 'dataset/' | |
MODEL_ID = 'roberta-base' | |
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, cache_dir=CACHE_DIR) | |
bea_dataset = load_dataset('json', | |
data_files={'train': [ROOT_DIR + "train_ds.json"], 'test': [ROOT_DIR + "test_ds.json"]}) | |
def set_seed(seed): | |
# REPRODUCIBILITY | |
random.seed(seed) | |
np.random.seed(seed) | |
torch.manual_seed(seed) | |
torch.backends.cudnn.deterministic = True | |
torch.backends.cudnn.benchmark = False | |
set_seed(42) | |
id2label = { | |
0: "Acquisition", | |
1: "Company_Invest", | |
2: "Contract", | |
3: "Government_Invest", | |
4: "Market_Outlook", | |
5: "New_Product", | |
6: "Other" | |
} | |
id2label = dict(sorted(id2label.items(), key=lambda item: item[1])) | |
label2id = {label: iid for iid, label in id2label.items()} | |
def preprocess_label2id_fn(example): | |
example["label"] = label2id[example["label"]] | |
return example | |
def preprocess_tokenize_fn(examples): | |
return tokenizer(examples["text"], truncation=True) | |
bea_dataset = bea_dataset.map(preprocess_label2id_fn, batched=False) | |
bea_dataset = bea_dataset.map(preprocess_tokenize_fn, batched=True) | |
data_collator = DataCollatorWithPadding(tokenizer=tokenizer) | |
accuracy = evaluate.load("accuracy") | |
def compute_metrics(eval_pred): | |
predictions, labels = eval_pred | |
predictions = np.argmax(predictions, axis=1) | |
return accuracy.compute(predictions=predictions, references=labels) | |
model = AutoModelForSequenceClassification.from_pretrained( | |
MODEL_ID, num_labels=len(id2label), id2label=id2label, label2id=label2id, cache_dir=CACHE_DIR, | |
) | |
training_args = TrainingArguments( | |
overwrite_output_dir=True, | |
output_dir="cache/logs/classifier", | |
learning_rate=5e-5, | |
gradient_accumulation_steps=4, | |
per_device_train_batch_size=16, | |
per_device_eval_batch_size=16, | |
num_train_epochs=10, | |
# weight_decay=0.01, | |
# evaluation_strategy="epoch", | |
# save_strategy="epoch", | |
# load_best_model_at_end=True, | |
) | |
trainer = Trainer( | |
model=model, | |
args=training_args, | |
train_dataset=bea_dataset["train"], | |
eval_dataset=bea_dataset["test"], | |
tokenizer=tokenizer, | |
data_collator=data_collator, | |
compute_metrics=compute_metrics, | |
) | |
trainer.train() | |
trainer.save_model('model/classifier/') | |
tokenizer.save_pretrained('model/classifier/') |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment