Clustering performance

[XifengGuo]

@hanxiao
I wonder what's the clustering performance of the state-of-the-art clustering algorithms on fashion-mnist.
I tested my algorithm and got an accuracy of 0.59 and NMI of 0.63.
Have you collected other clustering results?
Thanks.

[hanxiao]

hello Xifeng, thanks for testing. unfortunately, I don't have other cluster benchmark for comparison. I may add a table if people start to benchmark clustering algorithms on this dataset. As for now, I can add your repo to the readme.md if you want.

[kashif]

@XifengGuo indeed I think the best one I saw was https://github.com/lmcinnes/umap but again I do not have any numbers

[XifengGuo]

＠hanxiao
Thanks, I'll keep investigating the new results of clustering algorithms on Fashion-Mnist.
I believe it will benefit unsupervised learning community if results of state-of-the-art clustering algorithms are collected.
I'm looking forward to it.

For now, I only tested the most popular deep clustering algorithm Unsupervised Deep Embedding for Clustering Analysis (DEC).
If you think it is appropriate, you can add my result to the readme.md as a start. I'll appreciate it.

clustering model	Preprocessing	Fashion performance	MNIST performance	Submitter	Code
DEC	Normalization (just scale pixel values to [0,1])	ACC=0.618; NMI=0.651	ACC=0.912; NMI=0.869	XifengGuo	https://github.com/XifengGuo/DEC-keras

[XifengGuo]

@kashif Thanks for bring https://github.com/lmcinnes/umap to me, it looks really amazing.