Memory Efficient Kernel Approximation

Authors : Marin BOUTHEMY

This code is an implementation of the Memory Efficient Kernel Approximation (MEKA) algorithm designed by Si Si & al..

To use it just run the main function and test it on the ijcnn1 dataset.

Requirements

The library has some requirements :

• Python 3
• Numpy
• Pandas

To install all this requirement you can run:

pip install -r requirements.txt

Then you can just run the main to get the meka algorithm working.

python main.py

Files structure

The library contains the following files:

• main.py -> Run the algorithm and create differents kernel matrices (based on MEKA, Nystrom and classic computation) and calculate the score for each of the matrix.
• meka.py -> Implementation of the MEKA algorithm, composed on the 3 steps.
• utils.py -> Functions such as the computation of gaussian kernel or the Nystrom approximation algorithm.