New releases of WEKA: 3.8.5 and 3.9.5

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

New releases of WEKA: 3.8.5 and 3.9.5

Eibe Frank-2
Administrator
Hi all,

A new release of WEKA was overdue, so just like last year, we have made one just in time for Christmas. The new stable version is WEKA 3.8.5 and downloads are available at 

  https://waikato.github.io/weka-wiki/downloading_weka

There is also a WEKA 3.9.5 release, but the codebase for this development branch has not yet diverged from the 3.8 version.

As was the case for the last release, the new release files for the three main platforms (Windows, macOS, and Linux) include ZuluFX 11, and WEKA will run on this Java Virtual Machine if the default way of starting the WEKA application is used. Of course, it is also possible to run WEKA on any other Java 8 or later JVM that may be installed on a system by explicitly invoking the java command at a command-line interface to run weka.jar, and as always, there is a platform-independent .zip file for both WEKA 3.8.5 and WEKA 3.9.5 that does not include a Java Virtual Machine.

This release is primarily a bug-fix release. Here is a partial list of changes (for more details, see https://www.cs.waikato.ac.nz/~ml/weka/CHANGELOG-3-8-5):

* Tidied up command-line option handling, e.g., superfluous options should now be detected more consistently.

* Several bug fixes in the estimators package. Apart from a change to rounding in KernelEstimator, the changes should not have any effect in existing classifiers, etc.

* Bug fix and performance optimisation in the RemoveNominalValues filter.

* Improved stability of the line search in Optimization.java, which seems to improve the robustness of non-linear conjugate gradient descent (e.g., in Logistiic).

* Added option to Logistic so that standardisation of predictor attributes can be turned off.

* Startup check in WekaPackageManager now honours the offline flag.

* Bug fix in RandomCommittee:resampling will now be performed only if the base learner does not implement WeightedInstancesHandler and the instance weights are not all equal.

* Speed improvements to NormalizedPolyKernel. Changed check for linear model in RegOptimizer and SMO so that it is now possible to run the NormalizedPolyKernel with exponent 1.

* A notification is now being popped up by ArffTableModel when the user tries to rename an attribute and the name already exists in another attribute.

* Bug fixes in DictionaryBuilder and FixedDictionaryStringToWordVector.

* Bug fixes in Sorter step and FlowRunner of KnowledgeFlow.

* Package Manager now cleans any pre-existing installed package when installing a package from a zip file.

* Fixed a bug in the KnowledeFlow and Workbench apps that can cause a null pointer exception under macOS when using look and feels that differ from Aqua.

* Updated jflex library .jar to 1.8.2 and regenerated the two Parser.java files in WEKA using this new version.

* Updated database URL in DatabaseUtils.props.mysql to include the parameter nullDatabaseMeansCurrent=true. This maintains the behaviour of returning current database metadata as seen in drivers prior to version 5.1.47.

* Added modifications to enable additional converters for a given file suffix (e.g., additional converters for CSV files that may be available in WEKA packages).

* Preprocess panel now unsets any class attribute after applying a filter or producing data via a data generator. Associate panel now always unsets the class attribute before running the associator.

* Many bug fixes in ClusterGenerator and related classes, particularly SubspaceCluster, which is now usable and produces meaningful output when "totally uniform" data generation is selected.

* Added GUI editor for Range objects.

As always, let us know if you encounter any problems. We don't have a native version of WEKA for Apple ARM chips yet, but there is an OpenJDK version for Apple ARM already, and the JavaFX port is hopefully not far off.

Cheers,
Eibe


_______________________________________________
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: New releases of WEKA: 3.8.5 and 3.9.5

CelestinoXP
I would like to contribute to further improve Weka.

As we all know how newer versions always bring good things, like bug fixes,
code improvements, speed improvements, better performance with less resource
consumption, etc ...

So, I would like to know if there are updates to update all the libraries
that weka uses ... does it bring benefits or not?

Could this be a goal, to use the most recent bookstores for the beneficiary
of everything I said before?

If what you said makes sense, it would be interesting that in addition to
the main weka application, all weka packages have also been revised in order
to update as their libraries. I noticed that there are packages that are
local, but as their libraries are very old and are probably not benefiting
from all the bug fixes and performance improvements.

Usually just take a look at the "lib" folder and check the versions used.

One of these packages is AutoWeka, which I think is excellent, but it is
semi-obsolete, it is one of the most interesting packages on weka and hasn't
received a single update in a long time. I've tried to contact any github,
but the answer is always the same, the programmer says he doesn't have time
and just wants pull requests, but I'm not a java programmer. see
(https://github.com/automl/autoweka/issues/79) and
(https://github.com/automl/autoweka/issues/87)

Sometimes, simple things can make a difference for the better, like support
for java 11, since weka has also adapted to java 11.

I would like to see gurus like: Eibe, Fracpete, or Mark Hall, giving an
opinion on the advantages of having updated "libs", or even taking the old
packages and updating them to more current versions.

That makes sense?



--
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: New releases of WEKA: 3.8.5 and 3.9.5

valerio jus
In reply to this post by Eibe Frank-2

Hi Eibe, 

Thank you for the update. Appreciate the effort. 
 
* Preprocess panel now unsets any class attribute after applying a filter or producing data via a data generator. Associate panel now always unsets the class attribute before running the associator.


Kindly I believe this opinion in release 3.8.5. Correct? 

I am not sure that I got your point correctly, by saying " Preprocess panel now unsets any class attribute after applying a filter ", how could we see this? I have the new version (3.8.5) but could not recognize it. 

In addition, for this option " Associate panel now always unsets the class attribute before running the associator ", this means the class will be ignored by default. Am I, right? If this is the case, then why after using "weather.nominal" data, then applied Apriori which produces set of rules and the class  (play) is in the output rules, which has to be not there if the Associate panel always unsets the class:

Example rules:

 1. outlook=overcast 4 ==> play=yes 4    <conf:(1)> lift:(1.56) lev:(0.1) [1] conv:(1.43)
 2. temperature=cool 4 ==> humidity=normal 4    <conf:(1)> lift:(2) lev:(0.14) [2] conv:(2)
 3. humidity=normal windy=FALSE 4 ==> play=yes 4    <conf:(1)> lift:(1.56) lev:(0.1) [1] conv:(1.43) 


Could you please clarify my doubts?

Thank you so much. 

Cheers, 
Valerio
 

* Many bug fixes in ClusterGenerator and related classes, particularly SubspaceCluster, which is now usable and produces meaningful output when "totally uniform" data generation is selected.

* Added GUI editor for Range objects.

As always, let us know if you encounter any problems. We don't have a native version of WEKA for Apple ARM chips yet, but there is an OpenJDK version for Apple ARM already, and the JavaFX port is hopefully not far off.

Cheers,
Eibe

_______________________________________________
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
Reply | Threaded
Open this post in threaded view
|

Re: New releases of WEKA: 3.8.5 and 3.9.5

Mark Hall
On 22 Dec 2020, 8:45 AM +1300, Valerio jus <[hidden email]>, wrote:

Hi Eibe, 

Thank you for the update. Appreciate the effort. 
 
* Preprocess panel now unsets any class attribute after applying a filter or producing data via a data generator. Associate panel now always unsets the class attribute before running the associator.


Kindly I believe this opinion in release 3.8.5. Correct? 

I am not sure that I got your point correctly, by saying " Preprocess panel now unsets any class attribute after applying a filter ", how could we see this? I have the new version (3.8.5) but could not recognize it. 


The drop-down box on the pre-process panel is intended for visualisation and for designating an attribute to consider as the class when applying supervised filters. Previously, if you applied a supervised filter in the pre-process panel, the class attribute would remain set in the filtered data (which is also made available in the other panels of the explorer). The classify and cluster panels have their own class drop-down box that overrides any pre-set class in the data. However, the associate panel does not. This was causing a problem when applying the FPGrowth association rule learner as it does not allow a class attribute to be set. Apriori, on the other hand, allows a class attribute to be set but ignores it in normal operation mode; and there is a separate option in Apriori to specify a class attribute when the output of class association rules (CAR) is turned on.

In addition, for this option " Associate panel now always unsets the class attribute before running the associator ", this means the class will be ignored by default. Am I, right? If this is the case, then why after using "weather.nominal" data, then applied Apriori which produces set of rules and the class  (play) is in the output rules, which has to be not there if the Associate panel always unsets the class:

No, unsetting the class just means that that attribute is treated like any other one. So, you would expect to see it in the output. Furthermore, given that Apriori always treated all attributes in the same way (regardless of class designation), its behaviour is unchanged. The change only affects FPGrowth, as it has a check for a designated class attribute.

Example rules:

 1. outlook=overcast 4 ==> play=yes 4    <conf:(1)> lift:(1.56) lev:(0.1) [1] conv:(1.43)
 2. temperature=cool 4 ==> humidity=normal 4    <conf:(1)> lift:(2) lev:(0.14) [2] conv:(2)
 3. humidity=normal windy=FALSE 4 ==> play=yes 4    <conf:(1)> lift:(1.56) lev:(0.1) [1] conv:(1.43) 


Could you please clarify my doubts?

Thank you so much. 

Cheers, 
Valerio


_______________________________________________
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: New releases of WEKA: 3.8.5 and 3.9.5

Mark Hall
In reply to this post by CelestinoXP
We do make an attempt to keep packages up-to-date with regards to libraries and third-party integrations. Eibe, has been tireless with keeping the RPlugin package up-to-date with respect to schemes provided by the MLR R package. I have recently introduced a new version of the distributedWekaSpark package that brings it up to date with Spark 3. I’m currently looking at releasing a new version of kerasZoo soon that supports the Keras API in TensorFlow 2. These things take time and effort. Eibe, Peter and myself are, at best, part time contributors to Weka these days.

Third-party packages, such as AutoWeka, are outside of our scope to maintain or develop. AutoWeka, in particular, is a pain because it includes an embedded version of Weka that is always out-of-date and causes issues with the main Weka distribution.

Felix - we will look to fold interruptible Weka into the main branches and support ML-Plan next year (I promise :-)).

Cheers, 
Mark.

On 21 Dec 2020, 11:08 PM +1300, CelestinoXP <[hidden email]>, wrote:
I would like to contribute to further improve Weka.

As we all know how newer versions always bring good things, like bug fixes,
code improvements, speed improvements, better performance with less resource
consumption, etc ...

So, I would like to know if there are updates to update all the libraries
that weka uses ... does it bring benefits or not?

Could this be a goal, to use the most recent bookstores for the beneficiary
of everything I said before?

If what you said makes sense, it would be interesting that in addition to
the main weka application, all weka packages have also been revised in order
to update as their libraries. I noticed that there are packages that are
local, but as their libraries are very old and are probably not benefiting
from all the bug fixes and performance improvements.

Usually just take a look at the "lib" folder and check the versions used.

One of these packages is AutoWeka, which I think is excellent, but it is
semi-obsolete, it is one of the most interesting packages on weka and hasn't
received a single update in a long time. I've tried to contact any github,
but the answer is always the same, the programmer says he doesn't have time
and just wants pull requests, but I'm not a java programmer. see
(https://github.com/automl/autoweka/issues/79) and
(https://github.com/automl/autoweka/issues/87)

Sometimes, simple things can make a difference for the better, like support
for java 11, since weka has also adapted to java 11.

I would like to see gurus like: Eibe, Fracpete, or Mark Hall, giving an
opinion on the advantages of having updated "libs", or even taking the old
packages and updating them to more current versions.

That makes sense?



--
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

_______________________________________________
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: New releases of WEKA: 3.8.5 and 3.9.5

Eibe Frank
To add to this, looking at


which lists all the releases of official WEKA packages made by Waikato folks through SourceForge, it appears that we haven't just been sitting on our hands.

The most important package maintained by us that is absent from this list on SourceForge is wekaDeeplearning4j. We have made a separate project for this package on GitHub:


This package has seen quite a bit of development this year! For example, there is now a new inference panel for the Explorer, see


and the model zoo for image classification and feature extraction from images has been greatly extended and now includes EfficientNet models, for example.

Cheers,
Eibe

On Tue, Dec 22, 2020 at 9:32 AM Mark Hall <[hidden email]> wrote:
We do make an attempt to keep packages up-to-date with regards to libraries and third-party integrations. Eibe, has been tireless with keeping the RPlugin package up-to-date with respect to schemes provided by the MLR R package. I have recently introduced a new version of the distributedWekaSpark package that brings it up to date with Spark 3. I’m currently looking at releasing a new version of kerasZoo soon that supports the Keras API in TensorFlow 2. These things take time and effort. Eibe, Peter and myself are, at best, part time contributors to Weka these days.

Third-party packages, such as AutoWeka, are outside of our scope to maintain or develop. AutoWeka, in particular, is a pain because it includes an embedded version of Weka that is always out-of-date and causes issues with the main Weka distribution.

Felix - we will look to fold interruptible Weka into the main branches and support ML-Plan next year (I promise :-)).

Cheers, 
Mark.

On 21 Dec 2020, 11:08 PM +1300, CelestinoXP <[hidden email]>, wrote:
I would like to contribute to further improve Weka.

As we all know how newer versions always bring good things, like bug fixes,
code improvements, speed improvements, better performance with less resource
consumption, etc ...

So, I would like to know if there are updates to update all the libraries
that weka uses ... does it bring benefits or not?

Could this be a goal, to use the most recent bookstores for the beneficiary
of everything I said before?

If what you said makes sense, it would be interesting that in addition to
the main weka application, all weka packages have also been revised in order
to update as their libraries. I noticed that there are packages that are
local, but as their libraries are very old and are probably not benefiting
from all the bug fixes and performance improvements.

Usually just take a look at the "lib" folder and check the versions used.

One of these packages is AutoWeka, which I think is excellent, but it is
semi-obsolete, it is one of the most interesting packages on weka and hasn't
received a single update in a long time. I've tried to contact any github,
but the answer is always the same, the programmer says he doesn't have time
and just wants pull requests, but I'm not a java programmer. see
(https://github.com/automl/autoweka/issues/79) and
(https://github.com/automl/autoweka/issues/87)

Sometimes, simple things can make a difference for the better, like support
for java 11, since weka has also adapted to java 11.

I would like to see gurus like: Eibe, Fracpete, or Mark Hall, giving an
opinion on the advantages of having updated "libs", or even taking the old
packages and updating them to more current versions.

That makes sense?



--
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
_______________________________________________
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
Reply | Threaded
Open this post in threaded view
|

Re: New releases of WEKA: 3.8.5 and 3.9.5

valerio jus
In reply to this post by Mark Hall
Thank you Mark for the prompt reply. 

Please let me share with you WEKA 3.8.5 version that I have and its result. 

1- I loaded the "segment-challenge" data set, here is the resulted figure:
2- I applied the ClassBalancer filter, here is the related figure:
As you can see, there are no changes after and before the ClassBalancer filter (i.e., the red rectangle showed that nothing happened after applying ClassBalancer). Is this expected or not?

However, is this the interface of the Associate tab in WEKA version 3.8.5 (the new update), or it is different?


Thank you once again.

Cheers, 
Valerio

 

On Tue, Dec 22, 2020 at 4:20 AM Mark Hall <[hidden email]> wrote:
On 22 Dec 2020, 8:45 AM +1300, Valerio jus <[hidden email]>, wrote:

Hi Eibe, 

Thank you for the update. Appreciate the effort. 
 
* Preprocess panel now unsets any class attribute after applying a filter or producing data via a data generator. Associate panel now always unsets the class attribute before running the associator.


Kindly I believe this opinion in release 3.8.5. Correct? 

I am not sure that I got your point correctly, by saying " Preprocess panel now unsets any class attribute after applying a filter ", how could we see this? I have the new version (3.8.5) but could not recognize it. 


The drop-down box on the pre-process panel is intended for visualisation and for designating an attribute to consider as the class when applying supervised filters. Previously, if you applied a supervised filter in the pre-process panel, the class attribute would remain set in the filtered data (which is also made available in the other panels of the explorer). The classify and cluster panels have their own class drop-down box that overrides any pre-set class in the data. However, the associate panel does not. This was causing a problem when applying the FPGrowth association rule learner as it does not allow a class attribute to be set. Apriori, on the other hand, allows a class attribute to be set but ignores it in normal operation mode; and there is a separate option in Apriori to specify a class attribute when the output of class association rules (CAR) is turned on.

In addition, for this option " Associate panel now always unsets the class attribute before running the associator ", this means the class will be ignored by default. Am I, right? If this is the case, then why after using "weather.nominal" data, then applied Apriori which produces set of rules and the class  (play) is in the output rules, which has to be not there if the Associate panel always unsets the class:

No, unsetting the class just means that that attribute is treated like any other one. So, you would expect to see it in the output. Furthermore, given that Apriori always treated all attributes in the same way (regardless of class designation), its behaviour is unchanged. The change only affects FPGrowth, as it has a check for a designated class attribute.

Example rules:

 1. outlook=overcast 4 ==> play=yes 4    <conf:(1)> lift:(1.56) lev:(0.1) [1] conv:(1.43)
 2. temperature=cool 4 ==> humidity=normal 4    <conf:(1)> lift:(2) lev:(0.14) [2] conv:(2)
 3. humidity=normal windy=FALSE 4 ==> play=yes 4    <conf:(1)> lift:(1.56) lev:(0.1) [1] conv:(1.43) 


Could you please clarify my doubts?

Thank you so much. 

Cheers, 
Valerio

_______________________________________________
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
Reply | Threaded
Open this post in threaded view
|

Re: New releases of WEKA: 3.8.5 and 3.9.5

Peter Reutemann
> Please let me share with you WEKA 3.8.5 version that I have and its result.
>
> 1- I loaded the "segment-challenge" data set, here is the resulted figure:
>
>  1-loading the segment-challenge.png
>
> 2- I applied the ClassBalancer filter, here is the related figure:
>
>  2-after applying ClassBalancer.png
>
> As you can see, there are no changes after and before the ClassBalancer filter (i.e., the red rectangle showed that nothing happened after applying ClassBalancer). Is this expected or not?

The segment challenge is already balanced and applying that filter
won't change that (you displayed the first attribute instead of the
class attribute in your first screenshot). If you load the vote
dataset, which has a 267 / 168 distribution, and apply the filter, you
will get a 217.5 / 217.5 instance weight distribution.

> However, is this the interface of the Associate tab in WEKA version 3.8.5 (the new update), or it is different?

Not quite sure what you mean with that question. The unsetting of the
class attribute happens behind the scenes. Apriori has an actual
option for the class attribute index that you can set (if there is a
class attribute present).

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: New releases of WEKA: 3.8.5 and 3.9.5

valerio jus

Hi Peter, 

Thank you for your helpful reply. 


On Mon, Dec 28, 2020 at 3:54 PM Peter Reutemann <[hidden email]> wrote:
> Please let me share with you WEKA 3.8.5 version that I have and its result.
>
> 1- I loaded the "segment-challenge" data set, here is the resulted figure:
>
>  1-loading the segment-challenge.png
>
> 2- I applied the ClassBalancer filter, here is the related figure:
>
>  2-after applying ClassBalancer.png
>
> As you can see, there are no changes after and before the ClassBalancer filter (i.e., the red rectangle showed that nothing happened after applying ClassBalancer). Is this expected or not?

The segment challenge is already balanced and applying that filter
won't change that (you displayed the first attribute instead of the
class attribute in your first screenshot). If you load the vote
dataset, which has a 267 / 168 distribution, and apply the filter, you
will get a 217.5 / 217.5 instance weight distribution.

 
> However, is this the interface of the Associate tab in WEKA version 3.8.5 (the new update), or it is different?

Not quite sure what you mean with that question. The unsetting of the
class attribute happens behind the scenes. Apriori has an actual
option for the class attribute index that you can set (if there is a
class attribute present).

I thought the interface is modified and an option like "Ignore attributes" is added to the Associate tab (similar to the Cluster tab) in the new release. 

Thank you. 

Cheers, 
Valerio  

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

_______________________________________________
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: New releases of WEKA: 3.8.5 and 3.9.5

Peter Reutemann
>> > However, is this the interface of the Associate tab in WEKA version 3.8.5 (the new update), or it is different?
>>
>> Not quite sure what you mean with that question. The unsetting of the
>> class attribute happens behind the scenes. Apriori has an actual
>> option for the class attribute index that you can set (if there is a
>> class attribute present).
>
>
> I thought the interface is modified and an option like "Ignore attributes" is added to the Associate tab (similar to the Cluster tab) in the new release.

For such cases you can use the FilteredAssociator in conjunction with
the Remove filter.

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: New releases of WEKA: 3.8.5 and 3.9.5

valerio jus
Thank you, Peter, for the well explanation.


Cheers,
Valerio 

On Tue, 29 Dec 2020, 4:37 am Peter Reutemann, <[hidden email]> wrote:
>> > However, is this the interface of the Associate tab in WEKA version 3.8.5 (the new update), or it is different?
>>
>> Not quite sure what you mean with that question. The unsetting of the
>> class attribute happens behind the scenes. Apriori has an actual
>> option for the class attribute index that you can set (if there is a
>> class attribute present).
>
>
> I thought the interface is modified and an option like "Ignore attributes" is added to the Associate tab (similar to the Cluster tab) in the new release.

For such cases you can use the FilteredAssociator in conjunction with
the Remove filter.

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

_______________________________________________
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