train.py

from datasets import load_dataset, Dataset
from transformers import AutoTokenizer, AutoModelForSequenceClassification, TrainingArguments, Trainer
from sklearn.model_selection import train_test_split
import torch

# Step 1: Load Dataset
dataset = load_dataset("ealvaradob/phishing-dataset", "combined_reduced", trust_remote_code=True)

# Step 2: Convert to Pandas and Split
df = dataset['train'].to_pandas()
train_df, test_df = train_test_split(df, test_size=0.2, random_state=42)

# Step 3: Convert Back to Hugging Face Dataset
train_dataset = Dataset.from_pandas(train_df, preserve_index=False)
test_dataset = Dataset.from_pandas(test_df, preserve_index=False)

# Step 4: Tokenizer Initialization
tokenizer = AutoTokenizer.from_pretrained("bert-large-uncased")

# Step 5: Preprocess Function
def preprocess_data(examples):
    # Use the correct column name for the text data
    return tokenizer(examples['text'], truncation=True, padding='max_length', max_length=512)

# Step 6: Tokenize the Dataset
tokenized_train = train_dataset.map(preprocess_data, batched=True)
tokenized_test = test_dataset.map(preprocess_data, batched=True)

# Remove unused columns and set format for PyTorch
tokenized_train = tokenized_train.remove_columns(['text'])
tokenized_test = tokenized_test.remove_columns(['text'])
tokenized_train.set_format("torch")
tokenized_test.set_format("torch")

# Step 7: Model Initialization
model = AutoModelForSequenceClassification.from_pretrained("bert-large-uncased", num_labels=2)

# Step 8: Training Arguments
training_args = TrainingArguments(
    evaluation_strategy="epoch",
    learning_rate=2e-5,
    per_device_train_batch_size=16,
    per_device_eval_batch_size=16,
    num_train_epochs=3,
    weight_decay=0.01,
    save_strategy="epoch",
    logging_steps=10,
)

# Step 9: Trainer Setup
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_train,
    eval_dataset=tokenized_test,
)

# Step 10: Train the Model
trainer.train()

# Step 11: Save the Model
model.save_pretrained("./phishing_model")
tokenizer.save_pretrained("./phishing_model")

# Step 12: Inference Example
# Load the saved model for inference
loaded_tokenizer = AutoTokenizer.from_pretrained("./phishing_model")
loaded_model = AutoModelForSequenceClassification.from_pretrained("./phishing_model")

# Example input
text = "Your account has been compromised, please reset your password now!"
inputs = loaded_tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=512)

# Run inference
loaded_model.eval()
with torch.no_grad():
    outputs = loaded_model(**inputs)
    prediction = torch.argmax(outputs.logits, dim=-1).item()

print(f"Predicted label: {prediction}")  # 0 = non-phishing, 1 = phishing