Pruning M5P

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

Pruning M5P

Saleh Shahinfar-2

Dear Eibe, 


I am training a model tree (M5P) on my date using experimenter interface. If I set unpruned option to TRUE it will give me 151 models and once I set it to FALSE it gives me one model with not much improvement in accuracy (which is basically a regular linear regression). given the fact that highly pruned models often generalize purely on unseen data I would like to have a model which is slightly pruned and not going from 151 model to 1 model only.

Is there any way that I can modify  that?


Regards, 

S. 



_______________________________________________
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: Pruning M5P

Eibe Frank-2
Administrator
What you can do is turn pruning off and then increase the minimum number of instances per leaf, along the lines of something like this:

  java weka.Run .M5P -N -t ~/datasets/numeric/housing.arff -M 100

Cheers,
Eibe

> On 2/08/2017, at 5:42 PM, Saleh Shahinfar <[hidden email]> wrote:
>
> Dear Eibe,
>
> I am training a model tree (M5P) on my date using experimenter interface. If I set unpruned option to TRUE it will give me 151 models and once I set it to FALSE it gives me one model with not much improvement in accuracy (which is basically a regular linear regression). given the fact that highly pruned models often generalize purely on unseen data I would like to have a model which is slightly pruned and not going from 151 model to 1 model only.
> Is there any way that I can modify  that?
>
> Regards,
> S.
>
> _______________________________________________
> 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: Pruning M5P

Saleh Shahinfar-2
Thanks. it solved my problem.
S.


________________________________________
From: [hidden email] <[hidden email]> on behalf of Eibe Frank <[hidden email]>
Sent: Thursday, August 3, 2017 10:02 AM
To: Weka machine learning workbench list.
Subject: Re: [Wekalist] Pruning M5P

What you can do is turn pruning off and then increase the minimum number of instances per leaf, along the lines of something like this:

  java weka.Run .M5P -N -t ~/datasets/numeric/housing.arff -M 100

Cheers,
Eibe

> On 2/08/2017, at 5:42 PM, Saleh Shahinfar <[hidden email]> wrote:
>
> Dear Eibe,
>
> I am training a model tree (M5P) on my date using experimenter interface. If I set unpruned option to TRUE it will give me 151 models and once I set it to FALSE it gives me one model with not much improvement in accuracy (which is basically a regular linear regression). given the fact that highly pruned models often generalize purely on unseen data I would like to have a model which is slightly pruned and not going from 151 model to 1 model only.
> Is there any way that I can modify  that?
>
> Regards,
> S.
>
> _______________________________________________
> 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...