approximate_predict on the data used to fit and membership_vector inconsistencies

[ainkov]

Hey there,

Thanks much for this great library.
Here is the problem I'm facing:

As you can see, after a fit, I get the items assigned to cluster 2 and pass them to approximate_predict and I was a bit surprised to see that some of them are declared outliers. Then I decided to try membership_vector , but again the results are kind of unexpected to me , as you can see the 4th element (that was declared outlier by approximate_predict) now it seems to have quite a high probability of being in cluster 1 or at least that's how I interpret it.
Am I missing something, or :/ ?

I'm using version 0.8.25 , python 3.6

EDIT:
I found this comment in another issue:

"Even when that arg is not used, the argmax of probabilities generated by soft_clusters is often not the same as the cluster that is actually assigned." #360

Thanks much and keep going with this project!