FW: WEKA

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

FW: WEKA

Jayaseelan Marimuthu

Hi

Currently I am using WEKA for my research. I am using Stacking (PLSR, SMOreg and Random Forest). The leave one out cross validation for 1500 instances very slow despite I am using Server with Xeon 24 Cores. Almost 36 hours with 531 attributes. Please advice how I can increase the speeds.

 

 

Jaya

 

 

 

 


_______________________________________________
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: FW: WEKA

Eibe Frank
PLSR and SMOreg are single-threaded implementations and do not support multi-threading. RandomForest does support multi-threading: you need to set the numExecutionSlots parameter to the number of threads that you want to use. By default, it only uses a single thread (i.e., one core).

Stacking supports multi-threading: the base classifiers can be built in separate threads. In your case, assuming the three classifiers you mention are used as base classifiers (but what is used as the meta classifier?), you could set numExecutionSlots in Stacking to three.

With 1500 instances, the default stratified 10-fold cross-validation is more suitable than LOO-CV. If you use 1500-fold cross-validation for performance evaluation together with the default 10-fold internal cross-validation in Stacking, each of three base classifiers will be trained 15,000 times on approximately 90% of the full training set. The meta classifier in Stacking will be trained 1,500 times.

You can use the WekaServer package to multi-thread the cross-validation used for evaluation in the Classify tab of the Explorer:


Note that this will multi-thread the cross-validation used for evaluation of your stacked classifier (the "external" cross-validation), not the internal cross-validation used in Stacking (which cannot currently be multi-threaded in WEKA: instead multi-threading in Stacking is based on building different base classifiers in different threads, as mentioned above).

If you use the WekaServer with 10 threads and turn on multi-threading with three threads in Stacking, you should be fully saturating your 24 core machine. Try it and take a look in the Windows Task Manager to monitor what happens.

Cheers,
Eibe

On Thu, Mar 4, 2021 at 10:37 PM Jayaseelan Marimuthu <[hidden email]> wrote:

Hi

Currently I am using WEKA for my research. I am using Stacking (PLSR, SMOreg and Random Forest). The leave one out cross validation for 1500 instances very slow despite I am using Server with Xeon 24 Cores. Almost 36 hours with 531 attributes. Please advice how I can increase the speeds.

 

 

Jaya

 

 

 

 

_______________________________________________
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

_______________________________________________
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