RandomMachine

Random ensemble learning library that extends gradient boosting by randomly sampling base learners from a pool of LightGBM, CatBoost, XGBoost, and arbitrary scikit-learn-compatible estimators for improved ensemble diversity.

Installation

pip install randommachine

Or install from source:

git clone https://github.com/ghiffaryr/randommachine.git
cd randommachine
pip install -e .

Quick Start

Regression

from randommachine import RandomLGBMRegressor
from sklearn.datasets import make_regression
from sklearn.model_selection import train_test_split

X, y = make_regression(n_samples=1000, n_features=20, random_state=42)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

model = RandomLGBMRegressor(num_iterations=20, learning_rate=0.5, random_state=42)
model.fit(X_train, y_train, X_eval=X_test, y_eval=y_test)

predictions = model.predict(X_test)

Classification

from randommachine import RandomCatBoostClassifier
from sklearn.datasets import make_classification

X, y = make_classification(n_samples=1000, n_features=20, random_state=42)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

model = RandomCatBoostClassifier(num_iterations=20, learning_rate=0.5)
model.fit(X_train, y_train, X_eval=X_test, y_eval=y_test)

predictions = model.predict(X_test)
probabilities = model.predict_proba(X_test)

Available Models

LightGBM-based:

• RandomLGBMRegressor - Regression with random LightGBM base learners
• RandomLGBMClassifier - Classification with random LightGBM base learners

CatBoost-based:

• RandomCatBoostRegressor - Regression with random CatBoost base learners
• RandomCatBoostClassifier - Classification with random CatBoost base learners

XGBoost-based:

• RandomXGBRegressor - Regression with random XGBoost base learners
• RandomXGBClassifier - Classification with random XGBoost base learners

Generic (user-defined pool):

• RandomRegressor - Mix any sklearn-compatible regressors with custom probabilities
• RandomClassifier - Mix any sklearn-compatible classifiers with custom probabilities

Tutorial

An interactive Jupyter notebook is available in the /docs folder:

• Tutorial - Getting started guide with performance comparison vs plain LightGBM, CatBoost, and XGBoost baselines

cd docs/
jupyter notebook tutorial.ipynb

The tutorial includes side-by-side comparisons showing RandomMachine's improvement over fixed-family baselines.

Development

Run tests:

make test          # Run all tests
make test-cov      # With coverage report

Format code:

make format        # Black formatting
make lint          # Flake8 linting

License

MIT License - see LICENSE