Meka (multi-label learning and evaluation.)

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

Meka (multi-label learning and evaluation.)

Michael Hall
Are Meka newbie questions ok here or would filing a GitHub issue be better?
_______________________________________________
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: Meka (multi-label learning and evaluation.)

Peter Reutemann
> Are Meka newbie questions ok here or would filing a GitHub issue be better?

MEKA has its own mailing list:
http://waikato.github.io/meka/documentation/#getting-help

But please be aware that the project hasn't seen much activity lately.

Cheers, Peter
--
Peter Reutemann
Dept. of Computer Science
University of Waikato, NZ
+64 (7) 858-5174
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: Meka (multi-label learning and evaluation.)

Michael Hall


> On Sep 8, 2020, at 8:43 PM, Peter Reutemann <[hidden email]> wrote:
>
>> Are Meka newbie questions ok here or would filing a GitHub issue be better?
>
> MEKA has its own mailing list:
> http://waikato.github.io/meka/documentation/#getting-help
>
> But please be aware that the project hasn't seen much activity lately.
>

I did see it had been a bit since a release. I noticed you replied to a GitHub issue a week or two back.
I’ll post to the Meka list.

_______________________________________________
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: Meka (multi-label learning and evaluation.)

Michael Hall


On Sep 8, 2020, at 8:47 PM, Michael Hall <[hidden email]> wrote:



On Sep 8, 2020, at 8:43 PM, Peter Reutemann <[hidden email]> wrote:

Are Meka newbie questions ok here or would filing a GitHub issue be better?

MEKA has its own mailing list:
http://waikato.github.io/meka/documentation/#getting-help

But please be aware that the project hasn't seen much activity lately.


I did see it had been a bit since a release. I noticed you replied to a GitHub issue a week or two back.
I’ll post to the Meka list.

My direct mail bounced unknown user and the subscribe link seems busted. 
I’ll post here. If you think I should make it a GitHub issue let me know.
__________
I was going to try out Meka on a Kaggle competition
Mechanism of Action (MoA)
https://www.kaggle.com/c/lish-moa/overview
You might need to be registered.
Anyhow it is a "a multi-label classification problem.”

I am new to Meka but running into some problems maybe someone could identify where I get off-track.

I downloaded the relevant data. A training dataset and a target one. 
I read them into R, joined them and used RWeka to write out an arff.
In the arff I marked all the attributes, quite a few, from the target dataset as classes and clicked the appropriate button.
This seemed to make these the first attributes. I used the NumericToNominal filter to avoid errors for numeric classes not handled.
I removed the id attribute.
However, I get exceptions when I then try to run classifiers. Like…

meka.gui.explorer.ClassifyTab
Evaluation failed (train/test split):
java.lang.NegativeArraySizeException: -1
at weka.classifiers.AbstractClassifier.makeCopies(AbstractClassifier.java:122)
at meka.classifiers.multilabel.BR.buildClassifier(BR.java:65)

For train/test split

Evaluation failed (train/test split):
java.lang.IndexOutOfBoundsException: Index -1 out of bounds for length 1081
at java.base/jdk.internal.util.Preconditions.outOfBounds(Preconditions.java:64)
at java.base/jdk.internal.util.Preconditions.outOfBoundsCheckIndex(Preconditions.java:70)
at java.base/jdk.internal.util.Preconditions.checkIndex(Preconditions.java:248)
at java.base/java.util.Objects.checkIndex(Objects.java:373)
at java.base/java.util.ArrayList.get(ArrayList.java:427)
at weka.core.Instances.attribute(Instances.java:401)
at weka.core.Instances.renameAttribute(Instances.java:1161)
at meka.classifiers.multilabel.MULAN.buildClassifier(MULAN.java:129)

For Mulan. Similar errors for cross-validation. 
If you click the options window the only thing that seems sort of off is Threshold is set to “PCut1”?

Any thoughts on what I'm missing?

_______________________________________________
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: Meka (multi-label learning and evaluation.)

Peter Reutemann
> I did see it had been a bit since a release. I noticed you replied to a GitHub issue a week or two back.
> I’ll post to the Meka list.

I've been involved in Meka from a technical point of view, not from a
research point of view.

> My direct mail bounced unknown user and the subscribe link seems busted.

Looks like sf.net quietly discarded mailing list support...

> I’ll post here. If you think I should make it a GitHub issue let me know.

In that case, a github issue would be better.

> __________
> I was going to try out Meka on a Kaggle competition
> Mechanism of Action (MoA)
> https://www.kaggle.com/c/lish-moa/overview
> You might need to be registered.
> Anyhow it is a "a multi-label classification problem.”
>
> I am new to Meka but running into some problems maybe someone could identify where I get off-track.
>
> I downloaded the relevant data. A training dataset and a target one.
> I read them into R, joined them and used RWeka to write out an arff.
> In the arff I marked all the attributes, quite a few, from the target dataset as classes and clicked the appropriate button.
> This seemed to make these the first attributes. I used the NumericToNominal filter to avoid errors for numeric classes not handled.
> I removed the id attribute.
> However, I get exceptions when I then try to run classifiers. Like…
>
> meka.gui.explorer.ClassifyTab
> Evaluation failed (train/test split):
> java.lang.NegativeArraySizeException: -1
> at weka.classifiers.AbstractClassifier.makeCopies(AbstractClassifier.java:122)
> at meka.classifiers.multilabel.BR.buildClassifier(BR.java:65)
>
> For train/test split
>
> Evaluation failed (train/test split):
> java.lang.IndexOutOfBoundsException: Index -1 out of bounds for length 1081
> at java.base/jdk.internal.util.Preconditions.outOfBounds(Preconditions.java:64)
> at java.base/jdk.internal.util.Preconditions.outOfBoundsCheckIndex(Preconditions.java:70)
> at java.base/jdk.internal.util.Preconditions.checkIndex(Preconditions.java:248)
> at java.base/java.util.Objects.checkIndex(Objects.java:373)
> at java.base/java.util.ArrayList.get(ArrayList.java:427)
> at weka.core.Instances.attribute(Instances.java:401)
> at weka.core.Instances.renameAttribute(Instances.java:1161)
> at meka.classifiers.multilabel.MULAN.buildClassifier(MULAN.java:129)
>
> For Mulan. Similar errors for cross-validation.
> If you click the options window the only thing that seems sort of off is Threshold is set to “PCut1”?
>
> Any thoughts on what I'm missing?

Would have to see dataset and classifier configuration before being
able to comment.

Please note, I'm not a Meka user.

Cheers, Peter
--
Peter Reutemann
Dept. of Computer Science
University of Waikato, NZ
+64 (7) 858-5174
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: Meka (multi-label learning and evaluation.)

Michael Hall


On Sep 8, 2020, at 10:01 PM, Peter Reutemann <[hidden email]> wrote:

I did see it had been a bit since a release. I noticed you replied to a GitHub issue a week or two back.
I’ll post to the Meka list.

I've been involved in Meka from a technical point of view, not from a
research point of view.

My direct mail bounced unknown user and the subscribe link seems busted.

Looks like sf.net quietly discarded mailing list support...

I’ll post here. If you think I should make it a GitHub issue let me know.

In that case, a github issue would be better.


Do that tomorrow. Thanks.



_______________________________________________
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