train.py

import pandas as pd
from sklearn.model_selection import train_test_split
from transformers import ElectraTokenizer, ElectraForSequenceClassification, Trainer, TrainingArguments
import torch
from datasets import Dataset
import wandb
from sklearn.metrics import precision_recall_fscore_support, accuracy_score

# Load dataset
data = pd.read_csv('sentences.csv')

# Split dataset into train and eval sets
train_df, eval_df = train_test_split(data, test_size=0.2, random_state=42)

# Convert to Hugging Face Dataset
train_dataset = Dataset.from_pandas(train_df)
eval_dataset = Dataset.from_pandas(eval_df)

# Initialize the tokenizer and model
model_name = 'classla/bcms-bertic'
tokenizer = ElectraTokenizer.from_pretrained(model_name)
model = ElectraForSequenceClassification.from_pretrained(model_name, num_labels=3)

# Tokenize the datasets
def tokenize_function(examples):
    return tokenizer(examples['text'], truncation=True, padding='max_length', max_length=128)

train_dataset = train_dataset.map(tokenize_function, batched=True)
eval_dataset = eval_dataset.map(tokenize_function, batched=True)

# Set format for PyTorch
train_dataset.set_format(type='torch', columns=['input_ids', 'attention_mask', 'label'])
eval_dataset.set_format(type='torch', columns=['input_ids', 'attention_mask', 'label'])

# Define the compute_metrics function
def compute_metrics(p):
    preds = p.predictions.argmax(-1)
    precision, recall, f1, _ = precision_recall_fscore_support(p.label_ids, preds, average='weighted')
    acc = accuracy_score(p.label_ids, preds)
    return {
        'accuracy': acc,
        'precision': precision,
        'recall': recall,
        'f1': f1
    }

# Define the training arguments
training_args = TrainingArguments(
    output_dir='./results',
    evaluation_strategy='epoch',
    save_strategy='epoch',
    learning_rate=1e-5,
    per_device_train_batch_size=128,
    per_device_eval_batch_size=128,
    num_train_epochs=20,
    weight_decay=0.01,
    warmup_steps=500,
    logging_dir='./logs',
    logging_steps=10,
    save_total_limit=20,
    load_best_model_at_end=True,
    metric_for_best_model='accuracy',
    report_to='wandb',
    run_name='sentiment-classification',
)

# Initialize WandB
wandb.init(project="sentiment-classification", entity="dejan")

# Define Trainer
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=train_dataset,
    eval_dataset=eval_dataset,
    compute_metrics=compute_metrics
)

# Train the model
trainer.train()

# Evaluate the model
trainer.evaluate()

# Finish the WandB run
wandb.finish()

# Save the model
model.save_pretrained('./sentiment-model')
tokenizer.save_pretrained('./sentiment-model')