Weighted clusters

[jeancroy]

The CKmean algorithm support weighted cluster.

• I'm interested in that functionality and will probably make a fork.

• The only difference seems to be at the level of dissimilarity evaluation. (And a few summaries, like cumulative sum of weights)

• They do have test for that case that I can port to your format.

• Would you be interested in that going back to main branch ? If so do you have requests / pointers ?

• Api would probably be a weight array that can be empty or omitted to collapse to the current api.

Example use:

I'm interested in compressing an histogram, where x would be bin positions and w would be bins heights.
A naive approach would be to repeat the x value a number of time proportional to w,
but I work with probability so I don't have integer count.
Also this just increase time and memory complexity for nothing.