training/train_sentiment.py

import argparse, pandas as pd
from datasets import Dataset
from transformers import AutoTokenizer, AutoModelForSequenceClassification, TrainingArguments, Trainer
from training.utils import compute_metrics_sentiment

parser = argparse.ArgumentParser()
parser.add_argument("--model_name", default="distilbert-base-uncased")
parser.add_argument("--train_csv", required=True)
parser.add_argument("--eval_csv", required=True)
parser.add_argument("--text_col", default="text")
parser.add_argument("--label_col", default="label")
parser.add_argument("--output_dir", default="./outputs/sentiment")
parser.add_argument("--epochs", type=int, default=3)
parser.add_argument("--batch_size", type=int, default=16)
parser.add_argument("--lr", type=float, default=5e-5)
args = parser.parse_args()

train_df = pd.read_csv(args.train_csv)
eval_df  = pd.read_csv(args.eval_csv)

label_names = sorted(train_df[args.label_col].unique().tolist())
label2id = {l:i for i,l in enumerate(label_names)}
id2label = {i:l for l,i in label2id.items()}

def encode(df):
    tok = tokenizer(df[args.text_col].tolist(), truncation=True, padding=True)
    tok["labels"] = [label2id[l] for l in df[args.label_col].tolist()]
    return tok

tokenizer = AutoTokenizer.from_pretrained(args.model_name)
train_ds = Dataset.from_pandas(train_df).map(encode, batched=True, remove_columns=train_df.columns)
eval_ds  = Dataset.from_pandas(eval_df).map(encode, batched=True, remove_columns=eval_df.columns)

model = AutoModelForSequenceClassification.from_pretrained(
    args.model_name, num_labels=len(label_names), id2label=id2label, label2id=label2id
)

training_args = TrainingArguments(
    output_dir=args.output_dir,
    evaluation_strategy="epoch",
    learning_rate=args.lr,
    per_device_train_batch_size=args.batch_size,
    per_device_eval_batch_size=args.batch_size,
    num_train_epochs=args.epochs,
    weight_decay=0.01,
    load_best_model_at_end=True,
    metric_for_best_model="accuracy",
)

trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=train_ds,
    eval_dataset=eval_ds,
    tokenizer=tokenizer,
    compute_metrics=compute_metrics_sentiment,
)

trainer.train()
trainer.save_model(args.output_dir)
tokenizer.save_pretrained(args.output_dir)