Not an issue. So much as a doubt.

[nicodv]

Not an issue. So much as a doubt.

[nicodv]

Are you using the same data you showed me earlier? That only had 3 unique numerical values, so the k-means algorithm might be smart enough to ignore your silly input of 6. The k-prototypes data has more unique numerical and categorical pairs.

[akash-suresh]

Yes the same data. It had only three numerical values.

Yes. Exactly my question. How can the algorithm wrap it's head around this?

Isn't it supposed to initialize 6 centroids and go on as traditional methods do? (The scikit-learn kmeans module btw)

[nicodv]

Not that surprising that it quickly scans the data for no. of unique values, no? That makes the clustering obvious, and no algorithm is necessary.

[akash-suresh]

Can this be implemented in the kprototypes as well?

[nicodv]

Good idea. See #20

[akash-suresh]

👍