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

DES-P #219

Closed jayahm closed 3 years ago

jayahm commented 4 years ago

Hi,

As far as I understood, DES-P is a probabilistic method, but why it is not included the probabilistic methods in this library?

Menelau commented 3 years ago

This method does not follow the same structure from the probabilistic ones. It does not use the potential function model in the region of competence steps. That is the reason it is out.

jayahm commented 3 years ago

So, it is still a probabilistic DS method right?

jayahm commented 3 years ago

Hi,

I would like to re-open this.

Upon checking, the code to compute the competence is:

competences = np.mean(self.DSEL_processed_[neighbors, :], axis=1)

This is measuring the accuracy right?

What I understood, it is based on probability.

Am I wrong somewhere?

Menelau commented 3 years ago

This method is based solely on accuracy. It does not use the classifier's probabilities.

Em ter., 13 de out. de 2020 às 23:48, jayahm notifications@github.com escreveu:

Hi,

I would like to re-open this.

Upon checking, the code to compute the competence is:

competences = np.mean(self.DSELprocessed[neighbors, :], axis=1)

This is measuring the accuracy right?

What I understood, it is based on probability.

Am I wrong somewhere?

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub https://github.com/scikit-learn-contrib/DESlib/issues/219#issuecomment-708138544, or unsubscribe https://github.com/notifications/unsubscribe-auth/AD6SFZEOPZSUASC55IZJSHLSKUNQ3ANCNFSM4RLAS72Q .

jayahm commented 3 years ago

But, based on your paper, it was written probabilistic?

Menelau commented 3 years ago

It's an error in the paper.

Em qua., 14 de out. de 2020 às 00:37, jayahm notifications@github.com escreveu:

But, based on your paper, it was written probabilistic?

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub https://github.com/scikit-learn-contrib/DESlib/issues/219#issuecomment-708151123, or unsubscribe https://github.com/notifications/unsubscribe-auth/AD6SFZD4DXJ5XZJWDVA5QWDSKUTIDANCNFSM4RLAS72Q .

jayahm commented 3 years ago

So, can I consider it as an accuracy-based method?

Menelau commented 3 years ago

Yes, it is an accuracy based method. It simply selects the base classifiers that have local accuracy higher than random predictions.

jayahm commented 3 years ago

I see. Thanks!

jayahm commented 3 years ago

How about DES-KL? Is it accuracy-based DES too?