omar 1.0

Open Multivariate Adaptive Regression

omar

omar (Open Multivariate Adaptive Regression) is a Python package for discovering localised, linear structure in complex, high-dimensional datasets. It implements a modernized version of the Multivariate Adaptive Regression Splines (MARS) algorithm with improved numerical efficiency, based on modern rank-one update strategies and optional Fortran acceleration.

Installation

From PyPI

pip install omar

From Source

See the Getting Started wiki page.

---

Quick Example

import numpy as np
from omar import OMAR

np.random.seed(0)

x = np.random.normal(2, 1, size=(1000, 3))
noise = np.random.normal(size=1000)
y = ((x[:, 0] + np.maximum(0, (x[:, 0] - 1)) +
      np.maximum(0, (x[:, 0] - 1)) * x[:, 1] +
      np.maximum(0, (x[:, 0] - 1)) * np.maximum(0, (x[:, 1] - 0.8))) +
     0.12 * noise)
model = OMAR(max_nbases=11)
model.find_bases(X, y)
print(model)

Documentation

Visit omar wiki

---

Contributing

We welcome contributions! See Contributing for code guidelines, tests, and CI workflow.

---

License

This project is licensed under the MIT License. See LICENSE.txt.