kmodes icon indicating copy to clipboard operation
kmodes copied to clipboard

Online clustering

Open OlanaMi opened this issue 9 years ago • 5 comments

Hi,

thanks a lot for the work here! Do you plan on adding support for online training? Similar to scikit's partial_fit()?

Best, Olana

OlanaMi avatar Jul 26 '16 13:07 OlanaMi

Haven't given that any thought yet, but that would probably be a valuable addition.

nicodv avatar Jul 27 '16 03:07 nicodv

👍 on this!

abunsen avatar Sep 23 '16 03:09 abunsen

Great library ! Online learning would definitely be a great addition.

elmadj avatar Aug 09 '18 16:08 elmadj

@nicodv If I were to contribute to this feature, can you elaborate a little bit on why this is difficult ?

elmadj avatar Aug 10 '18 15:08 elmadj

@elmadj , we'd have to do something along the lines described here, where we add a partial_fit method: http://scikit-learn.org/stable/modules/scaling_strategies.html#incremental-learning

That would require a mini-batch implementation of k-modes, along the lines of scikit-learn's MiniBatchKMeans: https://github.com/scikit-learn/scikit-learn/blob/0.19.2/sklearn/cluster/k_means_.py#L1216

Seems to me quite a bit of work, but a worthwhile addition for sure.

nicodv avatar Aug 10 '18 16:08 nicodv