sigma_faust

Code to reproduce experiments in preprint Learning a sum of sparse matrix products by Moujahid Bou-Laouz, Valentin Emiya, Liva Ralaivola and Caroline Chaux (currently under review).

Install

Install requirements by creating an anaconda environment with Python 3.7 or 3.9:

conda create -n sigma_faust python=3.7 pandas numpy matplotlib scipy scikit-learn sympy openpyxl

or

conda create -n sigma_faust python=3.9 pandas numpy matplotlib scipy scikit-learn sympy openpyxl

Then activate the environment and install pyfaust:

conda activate sigma_faust

pip install pyfaust

Download and extract 'sigma_faust' files, go to the source code root directory and run

pip install -e .

Running the code

To reproduce the first experiment and the top two plots of Figure 1 (target matrix is composed of random entries), execute file exp_icassp_random.py to run the experiment and execute function plot_random(N=256) from sigma_faust.plot_icassp_results module to display the results.

To reproduce the second experiment, the bottom plot of Figure 1 and Figure 2 (target matrix is composed of data from MNIST),

• download and locally extract the data from the UCI repository
• run file extract_matrices_from_datasets.py after adapting path variable root_dir depending on the location of your data
• run file exp_icassp_data_matrices.py.
• execute function plot_digits() from sigma_faust.plot_icassp_results module to display the results.

Bug report

This code is in a preliminary version. Please report any bug to the authors.