does the algorithm work properly?

[onofricamila]

Hi! I was trying to understand the algorithm. It seemed to me that ...

• the number of 'q' micro clusters is not restricted as stated in the original paper

• to try the algorithm, you have to first call the partial fit method (online stage) and then call predict (offline stage)

However, many errors appeared when I tried to cluster a scikitlearn data set. I would like to clarify the points mentioned above. Thanks in advance!

[instance01]

Hm, I got it working the following way: First fit with nb_initial_points, then you can partial_fit online. I thought nb_micro_cluster is q, but I may be wrong. Haven't dug too deep into the code.

On a small toy dataset it seemed to work fine for me.