Locally weighted learning for WEKA

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

Locally weighted learning for WEKA

Anatoliy
Hello all!
Do I understand correctly - the "Locally weighted learning" method for
regression evens out the difference in the number of instances for intervals
of values by assigning certain weights to them? Where can these values of
the assigned weights be seen after the LWL operation? And yet - for
regression methods that are based on Gaussian distributions ( Gaussian
processes for regression without hyperparameter-tuning) - the LWL method
does not work, since it fits the uniform distribution. Or I'm wrong?

regards
Anatoliy/



--
Sent from: https://weka.8497.n7.nabble.com/
_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
To unsubscribe send an email to [hidden email]
To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|

Re: Locally weighted learning for WEKA

Peter Reutemann
> Do I understand correctly - the "Locally weighted learning" method for
> regression evens out the difference in the number of instances for intervals
> of values by assigning certain weights to them?

At prediction time, the neighborhood gets determined on the fly and
weights for the instances of that subset calculated and set. The base
classifier then gets trained with that dataset. Whether the classifier
takes advantage of the weights depends on whether it implements the
WeightedInstancesHandler interface:
https://weka.sourceforge.io/doc.dev/weka/core/WeightedInstancesHandler.html

> Where can these values of
> the assigned weights be seen after the LWL operation?

You can't, it happens behind the scenes inside LWL.

> And yet - for
> regression methods that are based on Gaussian distributions ( Gaussian
> processes for regression without hyperparameter-tuning) - the LWL method
> does not work, since it fits the uniform distribution. Or I'm wrong?

When GaussianProcesses became "weight-aware" I found that the
performance got worse... Not sure why.

Cheers, Peter
--
Peter Reutemann
Dept. of Computer Science
University of Waikato, NZ
+64 (7) 577-5304
http://www.cms.waikato.ac.nz/~fracpete/
http://www.data-mining.co.nz/
_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
To unsubscribe send an email to [hidden email]
To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|

Re: Locally weighted learning for WEKA

Anatoliy
Ok, thank you very much

regards
Anatoliy



--
Sent from: https://weka.8497.n7.nabble.com/
_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
To unsubscribe send an email to [hidden email]
To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html