search

scikit-learn-contrib / DESlib

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

META-DES.o #147

Open polidore opened 5 years ago

polidore commented 5 years ago

Is it possible to get the behavior in this paper using your META-DES and pyswarms:

https://arxiv.org/pdf/1811.00217.pdf

Thanks

Menelau commented 5 years ago

Hello,

I checked pyswarms library and although they do not offer the same BPSO implementation (with the custom V and S-shaped transfer functions) used in the article. However, it could be used as backend and customized in order to have the same implementation as in the paper.

One important thing is, the DESlib implementation of META-DES only using the five sets of meta-features that was defined in the original version of the META-DES framework. The META-DES.O added some new meta-features, so another thing that would need to be done is to add the new sets of meta-features. This should not be very difficult since they are mainly based on the probabilistic functions which are already implemented in the utils/prob_functions.py file.

We have plans to add this method in the next version of the lib.

gweisang commented 4 years ago

Hi, I would also be interested in the implementation of META-DES.O. Any idea when this would be added?

I could possibly lend a hand in developing this if need be.

Menelau commented 4 years ago

@gweisang

Hello, at the moment I'm working on the BPSO implementation which should be done probably tomorrow, and the additional stuffs to like global validation to prevent overfitting in the selection stage. With that we will have almost all components necessary to finish the META-DES.Oracle given that computing the majority of the additional meta-features can be done calling the functions from utils.prob_functions module.

It would be very good if you are to help, let me know in which part you want or can.