meta-heuristic algorithms in parameter settings

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

meta-heuristic algorithms in parameter settings

asadbtk
Hello to all

As there are some algorithms in Weka (i.e. Grid Search, Multi Search) which could be used in identifying the best parameters of a classifier i.e. SVM. I wonder can we use the meta-heuristic algorithms to set the parameters of the classifiers in Weka? I know there are some meta-heuristic algorithms but those are mainly available for feature selections. Can we use them as parameter settings?

_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
To subscribe, unsubscribe, etc., visit 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
|

Re: meta-heuristic algorithms in parameter settings

Eibe Frank-2
Administrator

There is Auto-WEKA, which uses Bayesian optimisation. A more recent alternative, not available as a WEKA package (yet),  is ML-Plan (https://fmohr.github.io/AILibs/projects/mlplan/).

 

Cheers,

Eibe

 

From: [hidden email]
Sent: Thursday, 25 July 2019 2:33 AM
To: [hidden email]
Subject: [Wekalist] meta-heuristic algorithms in parameter settings

 

Hello to all

 

As there are some algorithms in Weka (i.e. Grid Search, Multi Search) which could be used in identifying the best parameters of a classifier i.e. SVM. I wonder can we use the meta-heuristic algorithms to set the parameters of the classifiers in Weka? I know there are some meta-heuristic algorithms but those are mainly available for feature selections. Can we use them as parameter settings?

 


_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
To subscribe, unsubscribe, etc., visit 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
|

Re: meta-heuristic algorithms in parameter settings

asadbtk
Hi Eibe, thanks a lot.

I used Auto-Weka a few weeks ago but it takes a long time. It took more than two hours to run and give the output. It becomes very difficult to use, particularly when we have to employ multiple algorithms. 

On Thu, Jul 25, 2019 at 7:06 AM Eibe Frank <[hidden email]> wrote:

There is Auto-WEKA, which uses Bayesian optimisation. A more recent alternative, not available as a WEKA package (yet),  is ML-Plan (https://fmohr.github.io/AILibs/projects/mlplan/).

 

Cheers,

Eibe

 

From: [hidden email]
Sent: Thursday, 25 July 2019 2:33 AM
To: [hidden email]
Subject: [Wekalist] meta-heuristic algorithms in parameter settings

 

Hello to all

 

As there are some algorithms in Weka (i.e. Grid Search, Multi Search) which could be used in identifying the best parameters of a classifier i.e. SVM. I wonder can we use the meta-heuristic algorithms to set the parameters of the classifiers in Weka? I know there are some meta-heuristic algorithms but those are mainly available for feature selections. Can we use them as parameter settings?

 

_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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
|

Re: meta-heuristic algorithms in parameter settings

asadbtk
Hello Eibe, following my message, I got this result from Auto Weka. I just load the dataset and perform the SMOT function and run it:  What are the values of "arguments"? Also, Doesn't Auto Weka suggest the feature selection algorithms for a particular dataset?
The output is below: 
 
Auto-WEKA result:
best classifier: weka.classifiers.trees.RandomForest
arguments: [-I, 10, -K, 0, -depth, 0]   //// I did not find these arguments in RF classifier
attribute search: null
attribute search arguments: []
attribute evaluation: null
attribute evaluation arguments: []

On Thu, Jul 25, 2019 at 10:17 AM javed khan <[hidden email]> wrote:
Hi Eibe, thanks a lot.

I used Auto-Weka a few weeks ago but it takes a long time. It took more than two hours to run and give the output. It becomes very difficult to use, particularly when we have to employ multiple algorithms. 

On Thu, Jul 25, 2019 at 7:06 AM Eibe Frank <[hidden email]> wrote:

There is Auto-WEKA, which uses Bayesian optimisation. A more recent alternative, not available as a WEKA package (yet),  is ML-Plan (https://fmohr.github.io/AILibs/projects/mlplan/).

 

Cheers,

Eibe

 

From: [hidden email]
Sent: Thursday, 25 July 2019 2:33 AM
To: [hidden email]
Subject: [Wekalist] meta-heuristic algorithms in parameter settings

 

Hello to all

 

As there are some algorithms in Weka (i.e. Grid Search, Multi Search) which could be used in identifying the best parameters of a classifier i.e. SVM. I wonder can we use the meta-heuristic algorithms to set the parameters of the classifiers in Weka? I know there are some meta-heuristic algorithms but those are mainly available for feature selections. Can we use them as parameter settings?

 

_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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
|

Re: meta-heuristic algorithms in parameter settings

Eibe Frank-3
Take a look at the Javadoc:


It looks like feature selection was not found to be necessary for your data (within the limited search that Auto-WEKA performs).

Cheers,
Eibe

On Thu, Jul 25, 2019 at 9:31 PM javed khan <[hidden email]> wrote:
Hello Eibe, following my message, I got this result from Auto Weka. I just load the dataset and perform the SMOT function and run it:  What are the values of "arguments"? Also, Doesn't Auto Weka suggest the feature selection algorithms for a particular dataset?
The output is below: 
 
Auto-WEKA result:
best classifier: weka.classifiers.trees.RandomForest
arguments: [-I, 10, -K, 0, -depth, 0]   //// I did not find these arguments in RF classifier
attribute search: null
attribute search arguments: []
attribute evaluation: null
attribute evaluation arguments: []

On Thu, Jul 25, 2019 at 10:17 AM javed khan <[hidden email]> wrote:
Hi Eibe, thanks a lot.

I used Auto-Weka a few weeks ago but it takes a long time. It took more than two hours to run and give the output. It becomes very difficult to use, particularly when we have to employ multiple algorithms. 

On Thu, Jul 25, 2019 at 7:06 AM Eibe Frank <[hidden email]> wrote:

There is Auto-WEKA, which uses Bayesian optimisation. A more recent alternative, not available as a WEKA package (yet),  is ML-Plan (https://fmohr.github.io/AILibs/projects/mlplan/).

 

Cheers,

Eibe

 

From: [hidden email]
Sent: Thursday, 25 July 2019 2:33 AM
To: [hidden email]
Subject: [Wekalist] meta-heuristic algorithms in parameter settings

 

Hello to all

 

As there are some algorithms in Weka (i.e. Grid Search, Multi Search) which could be used in identifying the best parameters of a classifier i.e. SVM. I wonder can we use the meta-heuristic algorithms to set the parameters of the classifiers in Weka? I know there are some meta-heuristic algorithms but those are mainly available for feature selections. Can we use them as parameter settings?

 

_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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
|

Re: meta-heuristic algorithms in parameter settings

asadbtk
Yes Eibe, I understand that in some cases the feature selection algorithms do not perform well. But does Auto Weka or any other algorithm in Weka provides the detail about the parameter settings of the (meta-heuristic) feature selection algorithms? For example, what should be the population size, iteration size etc?


On Thu, Jul 25, 2019 at 12:46 PM Eibe Frank <[hidden email]> wrote:
Take a look at the Javadoc:


It looks like feature selection was not found to be necessary for your data (within the limited search that Auto-WEKA performs).

Cheers,
Eibe

On Thu, Jul 25, 2019 at 9:31 PM javed khan <[hidden email]> wrote:
Hello Eibe, following my message, I got this result from Auto Weka. I just load the dataset and perform the SMOT function and run it:  What are the values of "arguments"? Also, Doesn't Auto Weka suggest the feature selection algorithms for a particular dataset?
The output is below: 
 
Auto-WEKA result:
best classifier: weka.classifiers.trees.RandomForest
arguments: [-I, 10, -K, 0, -depth, 0]   //// I did not find these arguments in RF classifier
attribute search: null
attribute search arguments: []
attribute evaluation: null
attribute evaluation arguments: []

On Thu, Jul 25, 2019 at 10:17 AM javed khan <[hidden email]> wrote:
Hi Eibe, thanks a lot.

I used Auto-Weka a few weeks ago but it takes a long time. It took more than two hours to run and give the output. It becomes very difficult to use, particularly when we have to employ multiple algorithms. 

On Thu, Jul 25, 2019 at 7:06 AM Eibe Frank <[hidden email]> wrote:

There is Auto-WEKA, which uses Bayesian optimisation. A more recent alternative, not available as a WEKA package (yet),  is ML-Plan (https://fmohr.github.io/AILibs/projects/mlplan/).

 

Cheers,

Eibe

 

From: [hidden email]
Sent: Thursday, 25 July 2019 2:33 AM
To: [hidden email]
Subject: [Wekalist] meta-heuristic algorithms in parameter settings

 

Hello to all

 

As there are some algorithms in Weka (i.e. Grid Search, Multi Search) which could be used in identifying the best parameters of a classifier i.e. SVM. I wonder can we use the meta-heuristic algorithms to set the parameters of the classifiers in Weka? I know there are some meta-heuristic algorithms but those are mainly available for feature selections. Can we use them as parameter settings?

 

_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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
|

Re: meta-heuristic algorithms in parameter settings

Eibe Frank-3
My two cents: considering the large number of parameters that most meta-heuristic algorithms have, and the time it takes to run these algorithms for one set of fixed parameter values, establishing good parameter values for these algorithms generally requires extreme amounts of patience.

Anyway, MultiSearch in WEKA (available as a separate package) enables you to tune any combination of parameters in a classifier (such as an AttributeSelectedClassifier). It should be possible to use it in this context as well.

Cheers,
Eibe

On Fri, Jul 26, 2019 at 8:19 PM javed khan <[hidden email]> wrote:
Yes Eibe, I understand that in some cases the feature selection algorithms do not perform well. But does Auto Weka or any other algorithm in Weka provides the detail about the parameter settings of the (meta-heuristic) feature selection algorithms? For example, what should be the population size, iteration size etc?


On Thu, Jul 25, 2019 at 12:46 PM Eibe Frank <[hidden email]> wrote:
Take a look at the Javadoc:


It looks like feature selection was not found to be necessary for your data (within the limited search that Auto-WEKA performs).

Cheers,
Eibe

On Thu, Jul 25, 2019 at 9:31 PM javed khan <[hidden email]> wrote:
Hello Eibe, following my message, I got this result from Auto Weka. I just load the dataset and perform the SMOT function and run it:  What are the values of "arguments"? Also, Doesn't Auto Weka suggest the feature selection algorithms for a particular dataset?
The output is below: 
 
Auto-WEKA result:
best classifier: weka.classifiers.trees.RandomForest
arguments: [-I, 10, -K, 0, -depth, 0]   //// I did not find these arguments in RF classifier
attribute search: null
attribute search arguments: []
attribute evaluation: null
attribute evaluation arguments: []

On Thu, Jul 25, 2019 at 10:17 AM javed khan <[hidden email]> wrote:
Hi Eibe, thanks a lot.

I used Auto-Weka a few weeks ago but it takes a long time. It took more than two hours to run and give the output. It becomes very difficult to use, particularly when we have to employ multiple algorithms. 

On Thu, Jul 25, 2019 at 7:06 AM Eibe Frank <[hidden email]> wrote:

There is Auto-WEKA, which uses Bayesian optimisation. A more recent alternative, not available as a WEKA package (yet),  is ML-Plan (https://fmohr.github.io/AILibs/projects/mlplan/).

 

Cheers,

Eibe

 

From: [hidden email]
Sent: Thursday, 25 July 2019 2:33 AM
To: [hidden email]
Subject: [Wekalist] meta-heuristic algorithms in parameter settings

 

Hello to all

 

As there are some algorithms in Weka (i.e. Grid Search, Multi Search) which could be used in identifying the best parameters of a classifier i.e. SVM. I wonder can we use the meta-heuristic algorithms to set the parameters of the classifiers in Weka? I know there are some meta-heuristic algorithms but those are mainly available for feature selections. Can we use them as parameter settings?

 

_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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
|

Re: meta-heuristic algorithms in parameter settings

asadbtk
Hello Eibe. 

This paper used Tabu search to tune the parameters of SVM and have claimed that they used Weka. http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.570.2992&rep=rep1&type=pdf

I know that Tabu search is used in Weka but only for the feature selection. Could you please explain how they used it or they might have used Weka API (I am still not sure if it supports Tabu search for parameter tuning)? 

Thank you

On Mon, Jul 29, 2019 at 10:24 AM Eibe Frank <[hidden email]> wrote:
My two cents: considering the large number of parameters that most meta-heuristic algorithms have, and the time it takes to run these algorithms for one set of fixed parameter values, establishing good parameter values for these algorithms generally requires extreme amounts of patience.

Anyway, MultiSearch in WEKA (available as a separate package) enables you to tune any combination of parameters in a classifier (such as an AttributeSelectedClassifier). It should be possible to use it in this context as well.

Cheers,
Eibe

On Fri, Jul 26, 2019 at 8:19 PM javed khan <[hidden email]> wrote:
Yes Eibe, I understand that in some cases the feature selection algorithms do not perform well. But does Auto Weka or any other algorithm in Weka provides the detail about the parameter settings of the (meta-heuristic) feature selection algorithms? For example, what should be the population size, iteration size etc?


On Thu, Jul 25, 2019 at 12:46 PM Eibe Frank <[hidden email]> wrote:
Take a look at the Javadoc:


It looks like feature selection was not found to be necessary for your data (within the limited search that Auto-WEKA performs).

Cheers,
Eibe

On Thu, Jul 25, 2019 at 9:31 PM javed khan <[hidden email]> wrote:
Hello Eibe, following my message, I got this result from Auto Weka. I just load the dataset and perform the SMOT function and run it:  What are the values of "arguments"? Also, Doesn't Auto Weka suggest the feature selection algorithms for a particular dataset?
The output is below: 
 
Auto-WEKA result:
best classifier: weka.classifiers.trees.RandomForest
arguments: [-I, 10, -K, 0, -depth, 0]   //// I did not find these arguments in RF classifier
attribute search: null
attribute search arguments: []
attribute evaluation: null
attribute evaluation arguments: []

On Thu, Jul 25, 2019 at 10:17 AM javed khan <[hidden email]> wrote:
Hi Eibe, thanks a lot.

I used Auto-Weka a few weeks ago but it takes a long time. It took more than two hours to run and give the output. It becomes very difficult to use, particularly when we have to employ multiple algorithms. 

On Thu, Jul 25, 2019 at 7:06 AM Eibe Frank <[hidden email]> wrote:

There is Auto-WEKA, which uses Bayesian optimisation. A more recent alternative, not available as a WEKA package (yet),  is ML-Plan (https://fmohr.github.io/AILibs/projects/mlplan/).

 

Cheers,

Eibe

 

From: [hidden email]
Sent: Thursday, 25 July 2019 2:33 AM
To: [hidden email]
Subject: [Wekalist] meta-heuristic algorithms in parameter settings

 

Hello to all

 

As there are some algorithms in Weka (i.e. Grid Search, Multi Search) which could be used in identifying the best parameters of a classifier i.e. SVM. I wonder can we use the meta-heuristic algorithms to set the parameters of the classifiers in Weka? I know there are some meta-heuristic algorithms but those are mainly available for feature selections. Can we use them as parameter settings?

 

_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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
|

Re: meta-heuristic algorithms in parameter settings

Eibe Frank-3
Looking at the article, it seems that they implemented the code for tabu search themselves in WEKA, but I could not find a public location for the code. Your best bet is probably to contact Anna Corazza, the first author of the paper:


Cheers,
Eibe

On Sat, Aug 10, 2019 at 8:22 AM javed khan <[hidden email]> wrote:
Hello Eibe. 

This paper used Tabu search to tune the parameters of SVM and have claimed that they used Weka. http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.570.2992&rep=rep1&type=pdf

I know that Tabu search is used in Weka but only for the feature selection. Could you please explain how they used it or they might have used Weka API (I am still not sure if it supports Tabu search for parameter tuning)? 

Thank you

On Mon, Jul 29, 2019 at 10:24 AM Eibe Frank <[hidden email]> wrote:
My two cents: considering the large number of parameters that most meta-heuristic algorithms have, and the time it takes to run these algorithms for one set of fixed parameter values, establishing good parameter values for these algorithms generally requires extreme amounts of patience.

Anyway, MultiSearch in WEKA (available as a separate package) enables you to tune any combination of parameters in a classifier (such as an AttributeSelectedClassifier). It should be possible to use it in this context as well.

Cheers,
Eibe

On Fri, Jul 26, 2019 at 8:19 PM javed khan <[hidden email]> wrote:
Yes Eibe, I understand that in some cases the feature selection algorithms do not perform well. But does Auto Weka or any other algorithm in Weka provides the detail about the parameter settings of the (meta-heuristic) feature selection algorithms? For example, what should be the population size, iteration size etc?


On Thu, Jul 25, 2019 at 12:46 PM Eibe Frank <[hidden email]> wrote:
Take a look at the Javadoc:


It looks like feature selection was not found to be necessary for your data (within the limited search that Auto-WEKA performs).

Cheers,
Eibe

On Thu, Jul 25, 2019 at 9:31 PM javed khan <[hidden email]> wrote:
Hello Eibe, following my message, I got this result from Auto Weka. I just load the dataset and perform the SMOT function and run it:  What are the values of "arguments"? Also, Doesn't Auto Weka suggest the feature selection algorithms for a particular dataset?
The output is below: 
 
Auto-WEKA result:
best classifier: weka.classifiers.trees.RandomForest
arguments: [-I, 10, -K, 0, -depth, 0]   //// I did not find these arguments in RF classifier
attribute search: null
attribute search arguments: []
attribute evaluation: null
attribute evaluation arguments: []

On Thu, Jul 25, 2019 at 10:17 AM javed khan <[hidden email]> wrote:
Hi Eibe, thanks a lot.

I used Auto-Weka a few weeks ago but it takes a long time. It took more than two hours to run and give the output. It becomes very difficult to use, particularly when we have to employ multiple algorithms. 

On Thu, Jul 25, 2019 at 7:06 AM Eibe Frank <[hidden email]> wrote:

There is Auto-WEKA, which uses Bayesian optimisation. A more recent alternative, not available as a WEKA package (yet),  is ML-Plan (https://fmohr.github.io/AILibs/projects/mlplan/).

 

Cheers,

Eibe

 

From: [hidden email]
Sent: Thursday, 25 July 2019 2:33 AM
To: [hidden email]
Subject: [Wekalist] meta-heuristic algorithms in parameter settings

 

Hello to all

 

As there are some algorithms in Weka (i.e. Grid Search, Multi Search) which could be used in identifying the best parameters of a classifier i.e. SVM. I wonder can we use the meta-heuristic algorithms to set the parameters of the classifiers in Weka? I know there are some meta-heuristic algorithms but those are mainly available for feature selections. Can we use them as parameter settings?

 

_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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
|

Re: meta-heuristic algorithms in parameter settings

asadbtk
Thanks Eibe,

I have sent an email 

On Sat, Aug 10, 2019 at 4:37 AM Eibe Frank <[hidden email]> wrote:
Looking at the article, it seems that they implemented the code for tabu search themselves in WEKA, but I could not find a public location for the code. Your best bet is probably to contact Anna Corazza, the first author of the paper:


Cheers,
Eibe

On Sat, Aug 10, 2019 at 8:22 AM javed khan <[hidden email]> wrote:
Hello Eibe. 

This paper used Tabu search to tune the parameters of SVM and have claimed that they used Weka. http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.570.2992&rep=rep1&type=pdf

I know that Tabu search is used in Weka but only for the feature selection. Could you please explain how they used it or they might have used Weka API (I am still not sure if it supports Tabu search for parameter tuning)? 

Thank you

On Mon, Jul 29, 2019 at 10:24 AM Eibe Frank <[hidden email]> wrote:
My two cents: considering the large number of parameters that most meta-heuristic algorithms have, and the time it takes to run these algorithms for one set of fixed parameter values, establishing good parameter values for these algorithms generally requires extreme amounts of patience.

Anyway, MultiSearch in WEKA (available as a separate package) enables you to tune any combination of parameters in a classifier (such as an AttributeSelectedClassifier). It should be possible to use it in this context as well.

Cheers,
Eibe

On Fri, Jul 26, 2019 at 8:19 PM javed khan <[hidden email]> wrote:
Yes Eibe, I understand that in some cases the feature selection algorithms do not perform well. But does Auto Weka or any other algorithm in Weka provides the detail about the parameter settings of the (meta-heuristic) feature selection algorithms? For example, what should be the population size, iteration size etc?


On Thu, Jul 25, 2019 at 12:46 PM Eibe Frank <[hidden email]> wrote:
Take a look at the Javadoc:


It looks like feature selection was not found to be necessary for your data (within the limited search that Auto-WEKA performs).

Cheers,
Eibe

On Thu, Jul 25, 2019 at 9:31 PM javed khan <[hidden email]> wrote:
Hello Eibe, following my message, I got this result from Auto Weka. I just load the dataset and perform the SMOT function and run it:  What are the values of "arguments"? Also, Doesn't Auto Weka suggest the feature selection algorithms for a particular dataset?
The output is below: 
 
Auto-WEKA result:
best classifier: weka.classifiers.trees.RandomForest
arguments: [-I, 10, -K, 0, -depth, 0]   //// I did not find these arguments in RF classifier
attribute search: null
attribute search arguments: []
attribute evaluation: null
attribute evaluation arguments: []

On Thu, Jul 25, 2019 at 10:17 AM javed khan <[hidden email]> wrote:
Hi Eibe, thanks a lot.

I used Auto-Weka a few weeks ago but it takes a long time. It took more than two hours to run and give the output. It becomes very difficult to use, particularly when we have to employ multiple algorithms. 

On Thu, Jul 25, 2019 at 7:06 AM Eibe Frank <[hidden email]> wrote:

There is Auto-WEKA, which uses Bayesian optimisation. A more recent alternative, not available as a WEKA package (yet),  is ML-Plan (https://fmohr.github.io/AILibs/projects/mlplan/).

 

Cheers,

Eibe

 

From: [hidden email]
Sent: Thursday, 25 July 2019 2:33 AM
To: [hidden email]
Subject: [Wekalist] meta-heuristic algorithms in parameter settings

 

Hello to all

 

As there are some algorithms in Weka (i.e. Grid Search, Multi Search) which could be used in identifying the best parameters of a classifier i.e. SVM. I wonder can we use the meta-heuristic algorithms to set the parameters of the classifiers in Weka? I know there are some meta-heuristic algorithms but those are mainly available for feature selections. Can we use them as parameter settings?

 

_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit 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]
To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html