Create training/train_sentiment.py

training/train_sentiment.py

@@ -0,0 +1,61 @@
+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)