Deep K Mean Clustering

This repository contains implementation of Deep k-Means: Jointly clustering with k-Means and learning representations

How to run

python3 train_deep_k_means.py

The script launches comparison of Scikit-learn's implementation of K-Means with Deep K-Means. Test task is to separate MNIST data on 10 clusters. Used metrics are clustering accuracy and normalized mutual info score.

Expected output:

K-means
   ACC:  0.5134333333333333
   NMI:  0.4905763104545929
Deep K-means
   ACC:  0.84575
   NMI:  0.7918360450505724

Where ACC stands for accuracy and NMI is for normalized mutual info.

Enviroment specifications

• tensorflow==2.3.1
• numpy==1.18.1
• scikit-learn==0.21.3
• scipy==1.4.1