vaganboost 0.7.7

A hybrid model combining VAE, GAN, and LightGBM for boosting performance in high-energy physics or data analysis tasks.

VaganBoost: Hybrid VAE-GAN + LightGBM for Advanced Classification 0.7.6

Introduction

VAGANBoost is a hybrid generative model combining Variational Autoencoders (VAE) and Generative Adversarial Networks (GAN) with boosting techniques to enhance high-energy gamma-ray analysis.

Outlines

• Implements cVAE+cGAN and cGAN+cVAE+RandomForest models
• Designed for high-energy physics applications
• Utilizes deep learning and gradient boosting techniques

Key Features

• Hybrid Architecture: Combines deep generative models with gradient boosting
• VAE-GAN Integration: Joint latent space learning for improved feature representation
• LightGBM Classifier: State-of-the-art gradient boosting for final classification
• Automatic Feature Fusion: Combines VAE latent features with GAN-generated features
• Visualization Tools: Built-in metrics visualization and feature analysis
• PyTorch Backend: GPU-accelerated training with seamless CUDA support

Key Features Table

Feature	Description	Benefit
VAE-GAN Fusion	Combines reconstruction power of VAEs with GANs' generative capabilities	Enhanced feature learning
LightGBM Integration	Gradient boosting on learned features	Superior classification performance
Automatic GPU Support	Seamless CUDA integration	Faster training on supported hardware
Dynamic Feature Fusion	Combines latent and generated features	Improved representation learning
Visualization Suite	Built-in metrics plotting	Easy model evaluation

Troubleshooting

Common Issues:

1. CUDA Out of Memory: Reduce batch size or input dimensions
2. Poor Classification Performance:
   • Increase VAE latent dimensions
   • Adjust GAN-LightGBM feature ratio
3. Training Instability:

   model = VaganBoost(
       ...,
       vae_kl_weight=0.5,  # Adjust KL loss weight
       gan_gp_weight=10.0  # Add gradient penalty
   )
   

Installation

Prerequisites

• Python 3.6+
• NVIDIA GPU (recommended) with CUDA 11.0+

Install via pip

pip install vaganboost

From source

git clone https://github.com/AliBavarchee/vaganboost.git
cd vaganboost
pip install -e .

Quick Start

Basic Usage

from vaganboost import VaganBoost, load_data, split_data, normalize_data

# Prepare data
X, y = load_data("data.csv", target_column="label")
X_train, X_test, y_train, y_test = split_data(X, y, test_size=0.2)
X_train_norm, X_test_norm = normalize_data(X_train, X_test)

# Initialize model
model = VaganBoost(
    vae_input_dim=X_train_norm.shape[1],
    vae_latent_dim=64,
    gan_input_dim=100,
    num_class=4,
    device="cuda"
)

# Train components
model.train_vae(X_train_norm, epochs=100)
model.train_gan(X_train_norm, epochs=50)
model.train_lgbm(X_train_norm, y_train)

# Evaluate
accuracy = model.evaluate(X_test_norm, y_test)
print(f"Test Accuracy: {accuracy:.2%}")

Advanced Configuration

# Custom LightGBM parameters
lgbm_params = {
    'objective': 'multiclass',
    'num_class': 4,
    'metric': 'multi_logloss',
    'num_leaves': 63,
    'learning_rate': 0.1,
    'feature_fraction': 0.7
}

model = VaganBoost(
    vae_input_dim=128,
    vae_latent_dim=64,
    gan_input_dim=100,
    num_class=4,
    lgbm_params=lgbm_params,
    device="cuda"
)

Documentation

Core Components

Module	Description
data_utils	Data loading, splitting, and normalization
models	VAE, GAN, and LightGBM implementations
train	Joint training procedures
utils	Visualization and evaluation tools

Dependencies

See requirements.txt for required packages.

License

This project is licensed under the MIT License.

Contact Ali Bavarchee - [e-mail]

Project Link: https://github.com/AliBavarchee/vaganboost

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

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

---

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