Initialization is using euclidean distance

[xinyangz]

spherecluster/spherecluster/spherical_kmeans.py

Lines 44 to 48 in 701b0b1

	centers = _init_centroids(
	X, n_clusters, init, random_state=random_state, x_squared_norms=x_squared_norms
	)
	if verbose:
	print("Initialization complete")

I might be getting this wrong, but the code here seems to be using initialization function from sklearn. This could cause issue since the kmeans++ initialization in sklearn is based on euclidean distance. It should be replaced with cosine distance.

[jasonlaska]

Thank you, I'll look into it.

[rtrad89]

May I ask what the status of this issue is? I looked into the commits log but to no avail.
I am currently using init="random" as a safer option, and would like to switch to k-means++ if this issue is resolved.

Many thanks!

[jasonlaska]

Hi @rtrad89 or @t103z I'm not sure how much this will actually matter since this is just used for intiailization (and sklearn has chosen not to provide a distance-configurable version of kmeans++).

The code for the initialization is here: https://github.com/scikit-learn/scikit-learn/blob/0.22.X/sklearn/cluster/_k_means.py#L41, do you want to take a stab at a PR to port it for cosine dist?