search

tslearn-team / tslearn

The machine learning toolkit for time series analysis in Python
https://tslearn.readthedocs.io
BSD 2-Clause "Simplified" License
2.92k stars 342 forks source link

Add LCSS as a distance metric in TimeSeriesKMeans #434

Open pushkar5586 opened 1 year ago

pushkar5586 commented 1 year ago

Hi,

Is there a plan to support Longest Common Subsequence (LCSS) as a distance/similarity metric in the TimeSeriesKMeans model?

Many thanks.

pushkar5586 commented 1 year ago

This page says:

"This is the algorithm at stake when invoking tslearn.clustering.TimeSeriesKMeans with metric="lcss""

but, upon instantiating an object of type tslearn.clustering.TimeSeriesKMeans with metric="lcss", I get an exception saying that the only supported metric values are "euclidean", "dtw" and "dtw-soft".

I looked into the code of tslearn.clustering.TimeSeriesKMeans and LCSS isn't supported in there.

Would appreciate if anyone could share a timeline for an upgraded version of tslear with LCSS supported in its clustering models.

AMDonati commented 1 year ago

Hello, I am following-up on this issue regarding the integration of 'lcss' metric in tslearn.clustering.TimeSeriesKMeans. I indeed found that this integration is mentioned in the tslearn documentation release, but has not been effectively integrated.

I think it will be a great addition, as partitioning clustering algorithms depends heavily on the distance metric used, and as LCSS has proved to be more performant on noisy time-series whose length can differ greatly.

Many thanks in advance for the answer regarding this issue,

Arabiteff commented 10 months ago

you can use the sktime TimeSeriesKMeans it supports the lcss metric https://www.sktime.net/en/stable/api_reference/auto_generated/sktime.clustering.k_means.TimeSeriesKMeans.html