yakovdan/manifestonmnist

Few Sample Feature Selection via Feature Manifold Learning

This repository contains an implementation of ManiFest algorithm, executed on the MNIST dataset.

Prerequisites

• Octave. To run the code, GNU Octave is required. GNU Octave is a free and open source syntax compatible alternative to Matlab. To install GNU Octave in a Linux environment, run: sudo apt install octave The code depends on manopt, a manifold optimization package. Is is available for download at: manopt Once downloaded, unzip the file wherever is convenient and run the script importmanopt from Octave.
• Python environment. To run the code, a python environment is recommended. A python environment can be created via venv or conda
  • venv Create an environment by calling python -m venv ManiFest in the git clone directory. Then, requirements can be installed using pip install -r requirements.txt
  • conda Create a conda environment: conda env create -f environment.yaml

Running the code

If not already activated, activate the environment. Change directory to the git clone directory. Run: python MultiClassMNIST.py

if VISUALIZE is true, illustrative figures will be saved to disk. score_{percentile}.png contains the ManiFest score vector for the particular chosen percentile as scale factor {class_id}_{percentile}.png contains the eigenvector with the largest eigenvalue for each class, for the particular chosen percentile