search

scikit-learn-contrib / DESlib

A Python library for dynamic classifier and ensemble selection
BSD 3-Clause "New" or "Revised" License
479 stars 106 forks source link

Why RRC requires the number of neighbours? #210

Closed jayahm closed 4 years ago

jayahm commented 4 years ago

Hi,

I realize RRC and other methods where the region of competence is based on potential function requires the number of k.

What I 8understood is that, they use the whole DSEL.

Why the class of these methods, we still need to set k?

Menelau commented 4 years ago

@jayahm Hello,

It is not required to set the k value on the RRC model at all. By default it uses the whole DSEL which is the original version of those techniques.

We just allow the user to have more a bit more flexibility and limit the number of data points used in the calculation since it can significantly reduce computational cost when dealing with large datasets. But that is completely optional.

jayahm commented 4 years ago

I see. I got it. Thank you.