Quantcast

ensemble classifier?

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

ensemble classifier?

Bradley E Harris

Hi, everyone.
I am looking for a particular sort of ensemble classifier, and I don't see it.

I am trying to identify examples that fit into Class A of my test data.
To do that, I'd like to take in the results of multiple classifiers, and if any one of them identifies the example as belonging to Class A I would like to have it counted as a true positive.
I don't know if there is a way to do this, but I haven't found it in the available ensembles.
Do I need to generate the prediction CSVs and then process them?

Thanking you in advance,
Brad


_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html

Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: ensemble classifier?

Eibe Frank-3
There might be a hacky way to do this by (a) wrapping each classifier in your ensemble into a CostSensitiveClassifier (default cost matrix), (b) using Vote with the averaging rule to create the ensemble, and (c) wrapping the whole ensemble into a ThresholdSelector with an appropriate manually-specified threshold for your Class A. The threshold will depend on the number of classes you have. If you have 10 classes, it should be a value slightly smaller than 0.1.

CostSensitiveClassifier would be used solely to turn a class probability distribution like 0.8, 0.1, 0.05, 0.05 into the discrete distribution 1, 0, 0, 0 (i.e., the class with maximum probability in the base classifier gets probability 1 and all the other classes get probability 0).

Note that I haven't tried this and it may not work. :-)

Cheers,
Eibe

On Thu, Jan 5, 2017 at 10:33 AM, Bradley E Harris <[hidden email]> wrote:

Hi, everyone.
I am looking for a particular sort of ensemble classifier, and I don't see it.

I am trying to identify examples that fit into Class A of my test data.
To do that, I'd like to take in the results of multiple classifiers, and if any one of them identifies the example as belonging to Class A I would like to have it counted as a true positive.
I don't know if there is a way to do this, but I haven't found it in the available ensembles.
Do I need to generate the prediction CSVs and then process them?

Thanking you in advance,
Brad


_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html



_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: ensemble classifier?

Bradley E Harris

Hi, Eibe.
Thanks for the suggestion. :) For a quick and dirty solution I generated the prediction CSV files and processed them. :)

I'll have to play around with your suggestion and see if it will match what I got from this technique.

Thanks,
Brad


Inactive hide details for Eibe Frank ---01/07/2017 03:55:07 AM---There might be a hacky way to do this by (a) wrapping each claEibe Frank ---01/07/2017 03:55:07 AM---There might be a hacky way to do this by (a) wrapping each classifier in your ensemble into a CostSe

From: Eibe Frank <[hidden email]>
To: "Weka machine learning workbench list." <[hidden email]>
Date: 01/07/2017 03:55 AM
Subject: Re: [Wekalist] ensemble classifier?
Sent by: [hidden email]





There might be a hacky way to do this by (a) wrapping each classifier in your ensemble into a CostSensitiveClassifier (default cost matrix), (b) using Vote with the averaging rule to create the ensemble, and (c) wrapping the whole ensemble into a ThresholdSelector with an appropriate manually-specified threshold for your Class A. The threshold will depend on the number of classes you have. If you have 10 classes, it should be a value slightly smaller than 0.1.

CostSensitiveClassifier would be used solely to turn a class probability distribution like 0.8, 0.1, 0.05, 0.05 into the discrete distribution 1, 0, 0, 0 (i.e., the class with maximum probability in the base classifier gets probability 1 and all the other classes get probability 0).

Note that I haven't tried this and it may not work. :-)

Cheers,
Eibe

On Thu, Jan 5, 2017 at 10:33 AM, Bradley E Harris <[hidden email]> wrote:
    Hi, everyone.
    I am looking for a particular sort of ensemble classifier, and I don't see it.

    I am trying to identify examples that fit into Class A of my test data.
    To do that, I'd like to take in the results of multiple classifiers, and if any one of them identifies the example as belonging to Class A I would like to have it counted as a true positive.
    I don't know if there is a way to do this, but I haven't found it in the available ensembles.
    Do I need to generate the prediction CSVs and then process them?

    Thanking you in advance,
    Brad


    _______________________________________________
    Wekalist mailing list
    Send posts to:
    [hidden email]
    List info and subscription status:
    https://list.waikato.ac.nz/mailman/listinfo/wekalist
    List etiquette:
    http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status:
https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette:
http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html




_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Loading...