🧰 mltune

mltune

Flexible ML hyperparameter tuning and feature selection toolkit
Supports scikit-learn models and optionally XGBoost / LightGBM.

✨ Features

• Auto-tune hyperparameters (GridSearch)
• Optional greedy backward feature elimination
• Modular Wrapper classes for scikit-learn, XGBoost, LightGBM
• Unit & integration tested, Python 3.12+
• Lightweight, simple API

📦 Installation

Install base (requires Python ≥3.8):

pip install mltune

For optional XGBoost / LightGBM support:

pip install mltune[xgboost,lgbm]

🚀 Example usage

from mltune.wrappers import RandomForestModelWrapper

# Load or prepare data
X, y, X_test = load_data()

# Initialize wrapper with all features
wrapper = RandomForestModelWrapper(features=list(X.columns))

# Auto-tune hyperparameters & feature set
wrapper.autotune(
    X, y,
    hyperparam_initial_info={
        'n_estimators': [90, 95, 100, 105, 110],
        'max_depth': [9, 10, 11]
    },
    feature_selection_strategy="greedy_backward",
    verbose=True,
    plot=True
)

# Wrapper will use calculated hyperparameters & feature set
predictions = wrapper.predict(X_test)

✅ Implemented

• Wrappers:
• RandomForestModelWrapper
• XGBoostModelWrapper
• LightGBMModelWrapper
• Auto hyperparameter tuning:
• grid_search
• Feature selection strategy:
• none (skip feature elimination)
• greedy_backward

🧭 Planned / Roadmap

• Add other feature selection strategies (e.g. forward, recursive)
• Add other hyperparameter tuning strategies (e.g. Bayesian optimization)
• Voting strategies

📦 Development

Clone repo, install dev deps:

uv pip install -e .[dev] --system

Run tests:

pytest -v

📚 Documentation

API Reference (HTML)

📜 License

Released under the MIT License.

📌 Status

Beta: Work in progress. Contributions and ideas welcome!