What do centroids mean and how are they computed?

[nicodv]

What do centroids mean and how are they computed?

[nicodv]

Your comment about decoding made me think... cluster_centroids_ now shows the decoded centroids with the original data labels, instead of letting the user figure that decoding out. :)

I enabled this by making cluster_centroids_ a computed property that decodes the internally used _enc_cluster_centroids (formerly known as cluster_centroids_) using _enc_map (both now 'private' properties).

[nicodv]

Internally, I encode all you color categories into integers for speed. That's why you see [[2, 0], [6, 2]] instead of color names. Have a look at kmodes_huang.enc_map_ for interpretation of those integers.

I don't use Euclidean distance in kmodes, but determine in how many dimension the colors match (matching dissimilarity). See the papers mentioned in README for details.

[BrazilForever11]

Thank you for clarification! Is there function to decode it quickly? It seems like scikit preprocessing.LabelEncode does something similar.