TypeError when using TimeSeriesKMeans from kmeans_plusplus requiring sample_weight argument

[inega]

I am currently having an issue with TimeSeriesKMeans (tslearn version 0.5.3.2, sklearn version 1.3.0, running on Windows, python 3.11). I initially tried to use the function on my own dataset, but managed to reproduce the error with the first example in the documentation page:

from tslearn.clustering import TimeSeriesKMeans
from tslearn.generators import random_walks

X = random_walks(n_ts=50, sz=32, d=1)
km = TimeSeriesKMeans(n_clusters=3, metric="euclidean", max_iter=5, random_state=0).fit(X)

Traceback (most recent call last):

  Cell In[1], line 6
    km = TimeSeriesKMeans(n_clusters=3, metric="euclidean", max_iter=5, random_state=0).fit(X)

  File ~\.conda\envs\subsenv\Lib\site-packages\tslearn\clustering\kmeans.py:780 in fit
    self._fit_one_init(X_, x_squared_norms, rs)

  File ~\.conda\envs\subsenv\Lib\site-packages\tslearn\clustering\kmeans.py:629 in _fit_one_init
    self.cluster_centers_ = _kmeans_plusplus(

TypeError: _kmeans_plusplus() missing 1 required positional argument: 'sample_weight'

I think this might be a version issue? The kmeans_pluplus docs indicate that the sample_weight argument was introduced in version 1.3 of scikit-learn, which was just released. I guess downgrading to a previous version will make it go away, but I thought of reporting it anyway, since other people will start getting the same error at some point.

[NimaSarajpoor]

tslearn/tslearn/clustering/kmeans.py

Lines 637 to 652 in f718dd5

	if SKLEARN_VERSION_GREATER_THAN_OR_EQUAL_TO_1_3_0:
	sample_weight = _check_sample_weight(None, X, dtype=X.dtype)
	self.cluster_centers_ = _kmeans_plusplus(
	X.reshape((n_ts, -1)),
	self.n_clusters,
	x_squared_norms=x_squared_norms,
	sample_weight=sample_weight,
	random_state=rs,
	)[0].reshape((-1, sz, d))
	else:
	self.cluster_centers_ = _kmeans_plusplus(
	X.reshape((n_ts, -1)),
	self.n_clusters,
	x_squared_norms=x_squared_norms,
	random_state=rs,
	)[0].reshape((-1, sz, d))

Hope it helps with the debugging process...

Apparently, this part is added recently.
See: a091483

I think the param sample_weights should be added to the .fit method. (see KernelKmeans for reference)