NMF

This repository shows the sample implementation of Euclidean-, I-divergence-, and IS-divergence-based NMF. The implementation visualizes the result of applying NMF to an observation matrix. You can use an arbitrary number of bases for the basis matrix.

Requirement

• docker: 20.10.21
• docker compose: v2.13.0

Usage

You can use the following command to run the demo.

git clone https://github.com/beginaid/NMF
cd NMF
docker compose up -d --build

All of the following options are managed by docker-compose.yml:

• L: Rows of the observation matrix
• N: Columns of the observation matrix
• M: Rows of the basis matrix
• n_ineration: Maximum number of updates
• divergence: "EU", "I", or "IS"
  • EU: Euclidean distance
  • I: I-divergence
  • IS: Itakura-Saito divergence

After the nmf container is up, a directory named "results" is created. The directory contains a graph describing the decreasing process of the divergence and a heatmap visualizing the observation and approximation matrices (shown at the beginning of README).

After you can see these outputs, delete the container resources using the following command.

docker compose down

Note that you can change the parameters in docker-compose.yml and run the demo as many times as you like before the resources are down.