mvulib 0.1.0.post1

A Python interface to Maximum Variance Unfolding

mvulib

Description

Maximum Variance Unfolding (MVU) dimensionality reduction algorithm implemented in MATLAB with use of SeDuMi (Self Dual Minimization) package for solving semidefinite programming.

Installation and Requirements

MATLAB Runtime 24.1 installation is required. It can be downloaded from the following address: MATLAB Runtime 24.1. Runtime must be added to the PATH upon installation. Supported Python versions are >=3.9, <=3.11. Installation of numpy, scipy and scikit-learn is required.

Usage

Package includes Mvu Python class that serves as an interface to MATLAB implementation. Upon installation the Mvu class is instantiated and operated in a way similar to that of other sklearn dimensionality reduction classes. For example:

from mvulib.mvu import Mvu
from sklearn.datasets import make_swiss_roll
X, t=make_swiss_roll(n_samples=1000, random_state=0)
mvu=Mvu(n_neighbors=6, angles=2) # angles=0 is much faster
Y=mvu.fit_transform(X, 2) # Two dimensional embedding.

References

Implementation is based on the following paper:

K.Q.Weinberger, L.K.Saul. "Unsupervised Learning of Image Manifolds by Semidefinite Programming".

Documentation

Documentation .ipynb notebooks can be obtained upon request. Contact e-mail address: [e-mail].

License

This project is licensed under the GNU General Public License v2.0 or later (GPL‑2.0‑or‑later).

The choice of this license is required because the project depends on SeDuMi, which is distributed under the GNU GPL. Any software that incorporates or links to SeDuMi must also be released under a GPL‑compatible license.

By using, modifying, or distributing this project, you agree to the terms of the GPL‑2.0‑or‑later.
A full copy of the license is provided in the LICENSE file included with this distribution.