Hi Igniters,

Currently, I'm working on adoption of clustering algorithms
(KMeans/FuzzyCMeans) to the new Partitioned Dataset.

KMeans was adopted without any troubles, but FuzzyCMeans couldn't be
adopted so easy

1. It uses local data structures to collect indices of rows presented in
dataset. It works with old matrix-style approach, but it doesn't work with
the new partitioned dataset (it supports close integration with Ignite
Cache and works with any types of data, not only matrices)

2. It doesn't predict fuzzy belonging to the vector of clusters. There is a
copy-paste apply() method from the KMeans and it's incorrect behaviour.

3. I found a few bugs with weighted coeffiecient recalculation.

Summary, algorithm could be adopted fastly and doesn't work correctly
according its specification.

I suggest to remove the source files in the current release and return in
2.6 with a few fixes.

What do you think?

Alexey Zinoviev

Reply via email to