striXhooT: Hybrid Generative Model for Regression Tasks

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

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

striXhooT

striXhooT is a hybrid machine learning package that integrates generative models (CVAE, CGAN) with LightGBM for enhanced predictive modeling. The package is designed for structured datasets where generating synthetic data can improve regression performance.

Features

• Train Conditional Variational Autoencoder (CVAE) and Conditional Generative Adversarial Network (CGAN) models.
• Use generative models to augment training data for better regression performance.
• Implement an optimized LightGBM model with optional hyperparameter tuning.
• Supports PCA and SVD for dimensionality reduction.
• Provides end-to-end orchestration of model training and evaluation.

Architecture

graph TD;
    A[Input Data] -->|Preprocessing| B[Feature Extraction]
    B -->|Dimensionality Reduction| C[PCA/SVD]
    C -->|Transformed Data| D{Generative Models}
    D -->|Train| E[Conditional VAE]
    D -->|Train| F[Conditional GAN]
    E -->|Generated Features| G[Augmented Data]
    F -->|Generated Features| G
    G -->|Train| H[LightGBM Regressor]
    H -->|Predictions| I[Final Output]

Loading

📌 Installation

To install striXhoot, clone the repository and install dependencies:

# Clone the repository
git clone https://github.com/AliBavarchee/strixhoot.git
cd strixhoot

# Install dependencies
pip install -r requirements.txt

Usage

Below is an example script to train the hybrid model using striXhoot:

import os
import argparse
from strixhoot.trainer import main as train_model

# Define paths
input_path = "input.csv"  # Path to the dataset
output_path = "output_results"  # Directory to save all models and results

# Define training configuration
config = {
    "input_path": input_path,
    "output_path": output_path,
    "seed": 42,  # Set a random seed for reproducibility
    "gen_model": "both",  # Train both CVAE and CGAN models
    "train_generative": True,  # Enable training of generative models
    "tune_lgbm": False,  # Set to True if hyperparameter tuning is needed
    "dim_reducer": "pca",  # Choose between "pca" or "svd"
    "n_iter": 20,  # Number of iterations for hyperparameter tuning (only if tune_lgbm=True)
}

# Ensure the output directory exists
os.makedirs(output_path, exist_ok=True)

# Convert dictionary to argparse namespace
args = argparse.Namespace(**config)

# Train the models using the defined configuration
train_model(args)

Configuration Options

Parameter	Description
input_path	Path to the dataset (CSV format)
output_path	Directory where results and models will be saved
seed	Random seed for reproducibility
gen_model	Choose between cvae, cgan, or both
train_generative	Boolean flag to enable generative model training
tune_lgbm	Boolean flag to enable LightGBM hyperparameter tuning
dim_reducer	Choose dimensionality reduction method: pca or svd
n_iter	Number of iterations for LightGBM hyperparameter tuning (only if tune_lgbm=True)

Results and Outputs

After running the script, the following artifacts will be saved in the output_results/ directory:

• Trained CVAE and CGAN models
• Trained LightGBM regressor
• Performance metrics and visualizations (such as true_vs_pred.png)

Contributor(s)

• Ali Bavarchee (ali.bavarchee@gmail.com)

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/ |