train_main.py

import torch
import torch.nn as nn
import pandas as pd
from model import LSTMModel
from preprocessing import preprocess_text
from data_loader import create_data_loader
from sklearn.model_selection import train_test_split
from sklearn.metrics import f1_score, roc_auc_score
from keras.preprocessing.text import Tokenizer
from keras_preprocessing.sequence import pad_sequences
import pickle
import train as tr
from torch.utils.data import Dataset, DataLoader
from data_loader import NewsDataset
import os

version = 9

if __name__ == "__main__":
    # fake_path = './data_1/Fake.csv'
    # true_path = './data_1/True.csv'
    # cleaned_path = './cleaned_news_data.csv'
    # # Load data
    # try:
    #     df = pd.read_csv(cleaned_path)
    #     df.dropna(inplace=True)
    #     print("Cleaned data found.")
    # except:
    #     print("No cleaned data found. Cleaning data now...")
    #     # Load the datasets
    #     true_news = pd.read_csv('data_1/True.csv')
    #     fake_news = pd.read_csv('data_1/Fake.csv')

    #     # Add labels
    #     true_news['label'] = 1
    #     fake_news['label'] = 0

    #     # Combine the datasets
    #     df = pd.concat([true_news, fake_news], ignore_index=True)

    #     # Drop unnecessary columns
    #     df.drop(columns=['subject', 'date'], inplace=True)

    #     df['title'] = df['title'].apply(preprocess_text)
    #     df['text'] = df['text'].apply(preprocess_text)

    #     df.to_csv('cleaned_news_data.csv', index=False)
    #     df.dropna(inplace=True)

    data_path = "./data_2/WELFake_Dataset.csv"
    cleaned_path = f"./output/version_{version}/cleaned_news_data_{version}.csv"
    # Load data
    try:
        df = pd.read_csv(cleaned_path)
        df.dropna(inplace=True)
        print("Cleaned data found.")
    except:
        print("No cleaned data found. Cleaning data now...")
        df = pd.read_csv(data_path)

        # Drop index
        df.drop(df.columns[0], axis=1, inplace=True)
        df.dropna(inplace=True)

        # Swapping labels around since it originally is the opposite
        df["label"] = df["label"].map({0: 1, 1: 0})

        df["title"] = df["title"].apply(preprocess_text)
        df["text"] = df["text"].apply(preprocess_text)

        # Create the directory if it does not exist
        os.makedirs(os.path.dirname(cleaned_path), exist_ok=True)
        df.to_csv(cleaned_path, index=False)
        print("Cleaned data saved.")

    # Splitting the data
    train_val, test = train_test_split(df, test_size=0.2, random_state=42)
    train, val = train_test_split(
        train_val, test_size=0.25, random_state=42
    )  # 0.25 * 0.8 = 0.2

    # Initialize the tokenizer
    tokenizer = Tokenizer()

    # Fit the tokenizer on the training data
    tokenizer.fit_on_texts(train["title"] + train["text"])

    with open(f"./output/version_{version}/tokenizer_{version}.pickle", "wb") as handle:
        pickle.dump(tokenizer, handle, protocol=pickle.HIGHEST_PROTOCOL)

    # Tokenize the data
    X_train_title = tokenizer.texts_to_sequences(train["title"])
    X_train_text = tokenizer.texts_to_sequences(train["text"])
    X_val_title = tokenizer.texts_to_sequences(val["title"])
    X_val_text = tokenizer.texts_to_sequences(val["text"])
    X_test_title = tokenizer.texts_to_sequences(test["title"])
    X_test_text = tokenizer.texts_to_sequences(test["text"])

    # Padding sequences
    max_length = 500
    X_train_title = pad_sequences(X_train_title, maxlen=max_length)
    X_train_text = pad_sequences(X_train_text, maxlen=max_length)
    X_val_title = pad_sequences(X_val_title, maxlen=max_length)
    X_val_text = pad_sequences(X_val_text, maxlen=max_length)
    X_test_title = pad_sequences(X_test_title, maxlen=max_length)
    X_test_text = pad_sequences(X_test_text, maxlen=max_length)

    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    print(f"Device: {device}")

    model = LSTMModel(len(tokenizer.word_index) + 1).to(device)

    # Convert data to PyTorch tensors
    train_data = NewsDataset(
        torch.tensor(X_train_title),
        torch.tensor(X_train_text),
        torch.tensor(train["label"].values),
    )
    val_data = NewsDataset(
        torch.tensor(X_val_title),
        torch.tensor(X_val_text),
        torch.tensor(val["label"].values),
    )
    test_data = NewsDataset(
        torch.tensor(X_test_title),
        torch.tensor(X_test_text),
        torch.tensor(test["label"].values),
    )

    train_loader = DataLoader(
        train_data,
        batch_size=32,
        shuffle=True,
        num_workers=6,
        pin_memory=True,
        persistent_workers=True,
    )
    val_loader = DataLoader(
        val_data,
        batch_size=32,
        shuffle=False,
        num_workers=6,
        pin_memory=True,
        persistent_workers=True,
    )
    test_loader = DataLoader(
        test_data,
        batch_size=32,
        shuffle=False,
        num_workers=6,
        pin_memory=True,
        persistent_workers=True,
    )

    criterion = nn.BCELoss()
    optimizer = torch.optim.Adam(model.parameters(), lr=0.001)

    trained_model, best_accuracy, best_epoch = tr.train(
        model=model,
        train_loader=train_loader,
        val_loader=val_loader,
        criterion=criterion,
        optimizer=optimizer,
        version=version,
        epochs=10,
        device=device,
        max_grad_norm=1.0,
        early_stopping_patience=3,
        early_stopping_delta=0.01,
    )

    print(f"Best model was saved at epoch: {best_epoch}")

    # Load the best model before testing
    best_model_path = f"./output/version_{version}/best_model_{version}.pth"
    model.load_state_dict(torch.load(best_model_path, map_location=device))

    # Testing
    model.eval()
    true_labels = []
    predicted_labels = []
    predicted_probs = []

    with torch.no_grad():
        correct = 0
        total = 0
        for titles, texts, labels in test_loader:
            titles, texts, labels = (
                titles.to(device),
                texts.to(device),
                labels.to(device).float(),
            )
            outputs = model(titles, texts).squeeze()

            predicted = (outputs > 0.5).float()
            total += labels.size(0)
            correct += (predicted == labels).sum().item()
            true_labels.extend(labels.cpu().numpy())
            predicted_labels.extend(predicted.cpu().numpy())
            predicted_probs.extend(outputs.cpu().numpy())

    test_accuracy = 100 * correct / total
    f1 = f1_score(true_labels, predicted_labels)
    auc_roc = roc_auc_score(true_labels, predicted_probs)

    print(
        f"Test Accuracy: {test_accuracy:.2f}%, F1 Score: {f1:.4f}, AUC-ROC: {auc_roc:.4f}"
    )

    # Create DataFrame and Save to CSV
    confusion_data = pd.DataFrame({"True": true_labels, "Predicted": predicted_labels})
    confusion_data.to_csv(
        f"./output/version_{version}/confusion_matrix_data_{version}.csv", index=False
    )