vaganboostktf 0.9.1

Hybrid VAE-GAN with LightGBM for class-imbalanced classification

vaganboostktf: VAE-GAN Boost by TF

=============================================

=============================================

Hybrid generative-classification framework combining Conditional VAE, Conditional GAN, and LightGBM for handling class-imbalanced classification tasks.

Features

• 🧬 Hybrid architecture combining generative and discriminative models
• ⚖️ Effective handling of class imbalance through synthetic data generation
• 🔄 Iterative training process with automatic model refinement
• 📊 Comprehensive evaluation metrics and visualizations
• 💾 Model persistence and reproducibility features
• 🖥️ Command-line interface for easy operation

Installation

pip install vaganboostktf

For development installation:

git clone https://github.com/yourusername/vaganboostktf.git
cd vaganboostktf
pip install -e .

Usage

Basic Python API

import math
import numpy as np
import pandas as pd
import joblib
from pathlib import Path
import dill
from vaganboostktf.data_preprocessor import DataPreprocessor
from vaganboostktf.cgan import CGAN
from vaganboostktf.lgbm_tuner import LightGBMTuner
from vaganboostktf.trainer import HybridModelTrainer
from vaganboostktf.utils import plot_confusion_matrix, plot_roc_curves, plot_pr_curves

# Load and prepare data
df = pd.read_csv("Input.csv")

# Identify features and target
feature_columns = [f"ClE{i}" for i in range(1, 26)]  # 25 features
target_column = "label"

# Initialize data preprocessor
preprocessor = DataPreprocessor()
X_train, X_test, y_train, y_test = preprocessor.prepare_data(
    df,
    feature_columns=[f"ClE{i}" for i in range(1, 26)],
    target_column="label"
)

# Prepare scaled datasets
X_train_scaled, X_test_scaled, y_train, y_test = preprocessor.prepare_data(
    df,
    feature_columns=feature_columns,
    target_column=target_column
)

# Initialize and train hybrid model
trainer = HybridModelTrainer(config={
    'num_classes': 4,
    'cvae_params': {
        'input_dim': 25,
        'latent_dim': 8,
        'num_classes': 4,
        'learning_rate': 0.01
    },
    'cgan_params': {
        'input_dim': 25,
        'latent_dim': 8,
        'num_classes': 4,
        'generator_lr': 0.0002,
        'discriminator_lr': 0.0002
    },
    'model_dir': 'trained_models',
    'cvae_epochs': 10,
    'cgan_epochs': 10,
    'lgbm_iterations': 10,
    'samples_per_class': 50
})
trainer.training_loop(
    X_train_scaled, y_train,
    X_test_scaled, y_test,
    iterations=5
)

print("Training completed! Best models saved in 'trained_models' directory")

# Visualization of the metrics and results

# Load trained LightGBM model
lgbm_model = joblib.load("trained_models/lgbm_model.pkl")

# Get predictions
y_pred = lgbm_model.predict(X_test_scaled)
y_proba = lgbm_model.predict_proba(X_test_scaled)

# Define class names
class_names = [str(i) for i in np.unique(y_test)]

# Plot and save confusion matrix
conf_matrix_fig = plot_confusion_matrix(y_test, y_pred, classes=class_names, normalize=True)
conf_matrix_fig.savefig("trained_models/confusion_matrix.png")

# Plot and save ROC curves
roc_curve_fig = plot_roc_curves(y_test, y_proba, classes=class_names)
roc_curve_fig.savefig("trained_models/roc_curves.png")

# Plot and save Precision-Recall curves
pr_curve_fig = plot_pr_curves(y_test, y_proba, classes=class_names)
pr_curve_fig.savefig("trained_models/pr_curves.png")

# Extract feature importances and corresponding feature names
feature_importance = lgbm_model.feature_importances_
feature_names = [f"ClE{i}" for i in range(1, 26)]  # 25 features

# Create a DataFrame for sorting
importance_df = pd.DataFrame({'Feature': feature_names, 'Importance': feature_importance})
importance_df = importance_df.sort_values(by='Importance', ascending=False).head(10)  # Top 10 features

# Plot feature importance
plt.figure(figsize=(10, 6))
sns.barplot(x='Importance', y='Feature', data=importance_df, palette='Blues_r')
plt.xlabel("Importance Score")
plt.ylabel("Feature")
plt.title("Top 10 Most Important Features")
plt.grid(axis='x', linestyle='--', alpha=0.6)

# Save and show plot
plt.savefig("trained_models/top_10_features.png", bbox_inches='tight')
plt.show()

Architecture

graph TD
    A[Input Data] --> B[Data Preprocessing]
    B --> C[CVAE Training]
    B --> D[CGAN Training]
    C --> E[Synthetic Data Generation]
    D --> E
    E --> F[Data Augmentation]
    F --> G[LightGBM Training]
    G --> H[Evaluation]
    H --> I[Model Persistence]

Key Components

• Conditional VAE: Generates class-conditioned synthetic samples
• Conditional GAN: Produces additional class-specific synthetic data
• LightGBM Tuner: Optimized gradient boosting with automated hyperparameter search
• Hybrid Trainer: Orchestrates iterative training process

Configuration

Default parameters can be modified through:

• Command-line arguments
• JSON configuration files
• Python API parameters

License

This project is licensed under the MIT License - see the LICENSE file for details.

=============================================

=============================================

| https://github.com/AliBavarchee/ |

| https://www.linkedin.com/in/ali-bavarchee-qip/ |