Romain Tavenard

@GillesVandewiele I think we do have such a parameter for `KernelKMeans` but not for `TimeSeriesKMeans`. And I agree that this would be the correct way to implement it. This should...

@Ninimama I understand your point, yet: 1. depending on the form of your weight computation function, I am not sure that the algorithm at stake in `TimeSeriesKMeans` would be guaranteed...

But for knn, weights are just used at predict time, they are not involved in any fit time optimization. Once again I feel that this could definitely break convergence which...

Hi @texaspandaa , Thanks for your interest in tslearn. Do you have a specific reference in mind for "shapelet clustering" ? I don't think I have ever heard of that...

Oh you're right, I forgot about this one. I tag this as a new feature request. If anyone is interested in working on this, do not hesitate to tell! Best,...

Hi @orko19 This is a very good question. In fact, this module is dedicated to temporal-only data but sure there are use cases in which part of the features are...

Hi @bfocassio This should definitely be investigated. I am already aware that our implementation induces a raise in memory usage when time series of different lengths are at stake (since...

The constraint set on `numpy` comes from `tensorflow`, which, to the best of my knowledge, requires numpy~=1.19, not more. However, our dependency on `tensorflow` is a soft one (only required...

Oh, if I'm correct, TF2.7 should be able to match with numpy1.21, so we might have a solution here...

Note that we are already able to provide wheels for tslearn 3.10, so for those who just want tslearn without the optional `tensorflow` dependency, it should work, I guess. This...