Hi,

The reuters-*21578 *dataset which is available at the weka homepage has all
the test and train arff files separated by categories. Is it possible to
make 1 big file with all the categories? Is there any restriction in doing
that?

Thanks

Hi,

The reuters-*21578 *dataset which is available at the weka homepage has all
the test and train arff files separated by categories. Is it possible to
make 1 big file with all the categories? Is there any restriction in doing
that?

Thanks


Hi,

there is one slight issue: the categories are NOT mutually exclusive,
i.e. this is not a multi-class problem, but rather it is a multi-label problem,
or in other words: a document can belong to more than one category.

cheers, Bernhard

On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
<> wrote:
> Hi,
> The reuters-21578  dataset which is available at the weka homepage has all
> the test and train arff files separated by categories. Is it possible to
> make 1 big file with all the categories? Is there any restriction in doing
> that?
> Thanks
> _______________________________________________
> Wekalist mailing list
> Send posts to:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>
>



--
---------------------------------------------------------------------
Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
http://www.cs.waikato.ac.nz/~bernhard                  +64 7 838 4041

_______________________________________________
Wekalist mailing list
Send posts to:
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
)

Hi,

The reuters-*21578 *dataset which is available at the weka homepage has all
the test and train arff files separated by categories. Is it possible to
make 1 big file with all the categories? Is there any restriction in doing
that?

Thanks


Hi,

there is one slight issue: the categories are NOT mutually exclusive,
i.e. this is not a multi-class problem, but rather it is a multi-label problem,
or in other words: a document can belong to more than one category.

cheers, Bernhard

On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
<> wrote:
> Hi,
> The reuters-21578  dataset which is available at the weka homepage has all
> the test and train arff files separated by categories. Is it possible to
> make 1 big file with all the categories? Is there any restriction in doing
> that?
> Thanks
> _______________________________________________
> Wekalist mailing list
> Send posts to:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>
>



--
---------------------------------------------------------------------
Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
http://www.cs.waikato.ac.nz/~bernhard                  +64 7 838 4041

_______________________________________________
Wekalist mailing list
Send posts to:
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
)
Hi Bernhard,

Yes you are right, it's a multi-label problem. If I'm not wrong, some
classifiers can handle multi-class problems (more than two class labels)
naturally, and others that are binary class learners need to be called for
each label (ie one-against-the-rest). This can me done using
MultiClassClassifier.

Am I right?

Which classifiers among SMO, NaiveBayes, J48 and kNN can handle multiclass
problems naturally?

2011/7/15 Bernhard Pfahringer <>

> Hi,
>
> there is one slight issue: the categories are NOT mutually exclusive,
> i.e. this is not a multi-class problem, but rather it is a multi-label
> problem,
> or in other words: a document can belong to more than one category.
>
> cheers, Bernhard
>
> On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
> <> wrote:
> > Hi,
> > The reuters-21578 dataset which is available at the weka homepage has
> all
> > the test and train arff files separated by categories. Is it possible to
> > make 1 big file with all the categories? Is there any restriction in
> doing
> > that?
> > Thanks
> > _______________________________________________
> > Wekalist mailing list
> > Send posts to:
> > List info and subscription status:
> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> > List etiquette:
> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> >
> >
>
>
>
> --
> ---------------------------------------------------------------------
> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
> http://www.cs.waikato.ac.nz/~bernhard +64 7 838 4041
>
> _______________________________________________
> Wekalist mailing list
> Send posts to:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>

Hi,

The reuters-*21578 *dataset which is available at the weka homepage has all
the test and train arff files separated by categories. Is it possible to
make 1 big file with all the categories? Is there any restriction in doing
that?

Thanks


Hi,

there is one slight issue: the categories are NOT mutually exclusive,
i.e. this is not a multi-class problem, but rather it is a multi-label problem,
or in other words: a document can belong to more than one category.

cheers, Bernhard

On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
<> wrote:
> Hi,
> The reuters-21578  dataset which is available at the weka homepage has all
> the test and train arff files separated by categories. Is it possible to
> make 1 big file with all the categories? Is there any restriction in doing
> that?
> Thanks
> _______________________________________________
> Wekalist mailing list
> Send posts to:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>
>



--
---------------------------------------------------------------------
Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
http://www.cs.waikato.ac.nz/~bernhard                  +64 7 838 4041

_______________________________________________
Wekalist mailing list
Send posts to:
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
)
Hi Bernhard,

Yes you are right, it's a multi-label problem. If I'm not wrong, some
classifiers can handle multi-class problems (more than two class labels)
naturally, and others that are binary class learners need to be called for
each label (ie one-against-the-rest). This can me done using
MultiClassClassifier.

Am I right?

Which classifiers among SMO, NaiveBayes, J48 and kNN can handle multiclass
problems naturally?

2011/7/15 Bernhard Pfahringer <>

> Hi,
>
> there is one slight issue: the categories are NOT mutually exclusive,
> i.e. this is not a multi-class problem, but rather it is a multi-label
> problem,
> or in other words: a document can belong to more than one category.
>
> cheers, Bernhard
>
> On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
> <> wrote:
> > Hi,
> > The reuters-21578 dataset which is available at the weka homepage has
> all
> > the test and train arff files separated by categories. Is it possible to
> > make 1 big file with all the categories? Is there any restriction in
> doing
> > that?
> > Thanks
> > _______________________________________________
> > Wekalist mailing list
> > Send posts to:
> > List info and subscription status:
> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> > List etiquette:
> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> >
> >
>
>
>
> --
> ---------------------------------------------------------------------
> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
> http://www.cs.waikato.ac.nz/~bernhard +64 7 838 4041
>
> _______________________________________________
> Wekalist mailing list
> Send posts to:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>


Hi,

multi-label != multi-class

As you say, most Weka algorithms handle multi-class,
and if not, there is always MultiClassClassifier.

Multi-label is a different story, you'd need to look at
packages like Meka or Mulan, among others:
http://meka.sourceforge.net/
http://mulan.sourceforge.net/

cheers, Bernhard

On Fri, Jul 15, 2011 at 12:19 PM, Diana Iris Calleja
<> wrote:
> Hi Bernhard,
> Yes you are right, it's a multi-label problem. If I'm not wrong, some
> classifiers can handle multi-class problems (more than two class labels)
> naturally, and others that are binary class learners need to be called for
> each label (ie one-against-the-rest). This can me done using
> MultiClassClassifier.
> Am I right?
> Which classifiers among SMO, NaiveBayes, J48 and kNN can handle multiclass
> problems naturally?
>
> 2011/7/15 Bernhard Pfahringer <>
>>
>> Hi,
>>
>> there is one slight issue: the categories are NOT mutually exclusive,
>> i.e. this is not a multi-class problem, but rather it is a multi-label
>> problem,
>> or in other words: a document can belong to more than one category.
>>
>> cheers, Bernhard
>>
>> On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
>> <> wrote:
>> > Hi,
>> > The reuters-21578  dataset which is available at the weka homepage has
>> > all
>> > the test and train arff files separated by categories. Is it possible to
>> > make 1 big file with all the categories? Is there any restriction in
>> > doing
>> > that?
>> > Thanks
>> > _______________________________________________
>> > Wekalist mailing list
>> > Send posts to:
>> > List info and subscription status:
>> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
>> > List etiquette:
>> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>> >
>> >
>>
>>
>>
>> --
>> ---------------------------------------------------------------------
>> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
>> http://www.cs.waikato.ac.nz/~bernhard                  +64 7 838 4041
>>
>> _______________________________________________
>> Wekalist mailing list
>> Send posts to:
>> List info and subscription status:
>> https://list.scms.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:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>
>



--
---------------------------------------------------------------------
Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
http://www.cs.waikato.ac.nz/~bernhard                  +64 7 838 4041

_______________________________________________
Wekalist mailing list
Send posts to:
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
)

Hi,

The reuters-*21578 *dataset which is available at the weka homepage has all
the test and train arff files separated by categories. Is it possible to
make 1 big file with all the categories? Is there any restriction in doing
that?

Thanks


Hi,

there is one slight issue: the categories are NOT mutually exclusive,
i.e. this is not a multi-class problem, but rather it is a multi-label problem,
or in other words: a document can belong to more than one category.

cheers, Bernhard

On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
<> wrote:
> Hi,
> The reuters-21578  dataset which is available at the weka homepage has all
> the test and train arff files separated by categories. Is it possible to
> make 1 big file with all the categories? Is there any restriction in doing
> that?
> Thanks
> _______________________________________________
> Wekalist mailing list
> Send posts to:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>
>



--
---------------------------------------------------------------------
Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
http://www.cs.waikato.ac.nz/~bernhard                  +64 7 838 4041

_______________________________________________
Wekalist mailing list
Send posts to:
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
)
Hi Bernhard,

Yes you are right, it's a multi-label problem. If I'm not wrong, some
classifiers can handle multi-class problems (more than two class labels)
naturally, and others that are binary class learners need to be called for
each label (ie one-against-the-rest). This can me done using
MultiClassClassifier.

Am I right?

Which classifiers among SMO, NaiveBayes, J48 and kNN can handle multiclass
problems naturally?

2011/7/15 Bernhard Pfahringer <>

> Hi,
>
> there is one slight issue: the categories are NOT mutually exclusive,
> i.e. this is not a multi-class problem, but rather it is a multi-label
> problem,
> or in other words: a document can belong to more than one category.
>
> cheers, Bernhard
>
> On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
> <> wrote:
> > Hi,
> > The reuters-21578 dataset which is available at the weka homepage has
> all
> > the test and train arff files separated by categories. Is it possible to
> > make 1 big file with all the categories? Is there any restriction in
> doing
> > that?
> > Thanks
> > _______________________________________________
> > Wekalist mailing list
> > Send posts to:
> > List info and subscription status:
> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> > List etiquette:
> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> >
> >
>
>
>
> --
> ---------------------------------------------------------------------
> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
> http://www.cs.waikato.ac.nz/~bernhard +64 7 838 4041
>
> _______________________________________________
> Wekalist mailing list
> Send posts to:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>


Hi,

multi-label != multi-class

As you say, most Weka algorithms handle multi-class,
and if not, there is always MultiClassClassifier.

Multi-label is a different story, you'd need to look at
packages like Meka or Mulan, among others:
http://meka.sourceforge.net/
http://mulan.sourceforge.net/

cheers, Bernhard

On Fri, Jul 15, 2011 at 12:19 PM, Diana Iris Calleja
<> wrote:
> Hi Bernhard,
> Yes you are right, it's a multi-label problem. If I'm not wrong, some
> classifiers can handle multi-class problems (more than two class labels)
> naturally, and others that are binary class learners need to be called for
> each label (ie one-against-the-rest). This can me done using
> MultiClassClassifier.
> Am I right?
> Which classifiers among SMO, NaiveBayes, J48 and kNN can handle multiclass
> problems naturally?
>
> 2011/7/15 Bernhard Pfahringer <>
>>
>> Hi,
>>
>> there is one slight issue: the categories are NOT mutually exclusive,
>> i.e. this is not a multi-class problem, but rather it is a multi-label
>> problem,
>> or in other words: a document can belong to more than one category.
>>
>> cheers, Bernhard
>>
>> On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
>> <> wrote:
>> > Hi,
>> > The reuters-21578  dataset which is available at the weka homepage has
>> > all
>> > the test and train arff files separated by categories. Is it possible to
>> > make 1 big file with all the categories? Is there any restriction in
>> > doing
>> > that?
>> > Thanks
>> > _______________________________________________
>> > Wekalist mailing list
>> > Send posts to:
>> > List info and subscription status:
>> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
>> > List etiquette:
>> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>> >
>> >
>>
>>
>>
>> --
>> ---------------------------------------------------------------------
>> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
>> http://www.cs.waikato.ac.nz/~bernhard                  +64 7 838 4041
>>
>> _______________________________________________
>> Wekalist mailing list
>> Send posts to:
>> List info and subscription status:
>> https://list.scms.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:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>
>



--
---------------------------------------------------------------------
Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
http://www.cs.waikato.ac.nz/~bernhard                  +64 7 838 4041

_______________________________________________
Wekalist mailing list
Send posts to:
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
)
Hi Bernhard,

I will try Meka package. I thought I could handle multi-label
with MultiClassClassifier because I read it in this forum:

http://forums.pentaho.com/showthread.php?52105-Overview-of-weka-classifiers-for-a-three-class-problem

So weka ONLY works with binary label cases. Is this right?

Many thanks, Diana.

2011/7/15 Bernhard Pfahringer <>

> Hi,
>
> multi-label != multi-class
>
> As you say, most Weka algorithms handle multi-class,
> and if not, there is always MultiClassClassifier.
>
> Multi-label is a different story, you'd need to look at
> packages like Meka or Mulan, among others:
> http://meka.sourceforge.net/
> http://mulan.sourceforge.net/
>
> cheers, Bernhard
>
> On Fri, Jul 15, 2011 at 12:19 PM, Diana Iris Calleja
> <> wrote:
> > Hi Bernhard,
> > Yes you are right, it's a multi-label problem. If I'm not wrong, some
> > classifiers can handle multi-class problems (more than two class labels)
> > naturally, and others that are binary class learners need to be called
> for
> > each label (ie one-against-the-rest). This can me done using
> > MultiClassClassifier.
> > Am I right?
> > Which classifiers among SMO, NaiveBayes, J48 and kNN can handle
> multiclass
> > problems naturally?
> >
> > 2011/7/15 Bernhard Pfahringer <>
> >>
> >> Hi,
> >>
> >> there is one slight issue: the categories are NOT mutually exclusive,
> >> i.e. this is not a multi-class problem, but rather it is a multi-label
> >> problem,
> >> or in other words: a document can belong to more than one category.
> >>
> >> cheers, Bernhard
> >>
> >> On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
> >> <> wrote:
> >> > Hi,
> >> > The reuters-21578 dataset which is available at the weka homepage has
> >> > all
> >> > the test and train arff files separated by categories. Is it possible
> to
> >> > make 1 big file with all the categories? Is there any restriction in
> >> > doing
> >> > that?
> >> > Thanks
> >> > _______________________________________________
> >> > Wekalist mailing list
> >> > Send posts to:
> >> > List info and subscription status:
> >> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> >> > List etiquette:
> >> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> >> >
> >> >
> >>
> >>
> >>
> >> --
> >> ---------------------------------------------------------------------
> >> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
> >> http://www.cs.waikato.ac.nz/~bernhard +64 7 838 4041
> >>
> >> _______________________________________________
> >> Wekalist mailing list
> >> Send posts to:
> >> List info and subscription status:
> >> https://list.scms.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:
> > List info and subscription status:
> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> > List etiquette:
> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> >
> >
>
>
>
> --
> ---------------------------------------------------------------------
> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
> http://www.cs.waikato.ac.nz/~bernhard +64 7 838 4041
>
> _______________________________________________
> Wekalist mailing list
> Send posts to:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>

Hi,

The reuters-*21578 *dataset which is available at the weka homepage has all
the test and train arff files separated by categories. Is it possible to
make 1 big file with all the categories? Is there any restriction in doing
that?

Thanks


Hi,

there is one slight issue: the categories are NOT mutually exclusive,
i.e. this is not a multi-class problem, but rather it is a multi-label problem,
or in other words: a document can belong to more than one category.

cheers, Bernhard

On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
<> wrote:
> Hi,
> The reuters-21578  dataset which is available at the weka homepage has all
> the test and train arff files separated by categories. Is it possible to
> make 1 big file with all the categories? Is there any restriction in doing
> that?
> Thanks
> _______________________________________________
> Wekalist mailing list
> Send posts to:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>
>



--
---------------------------------------------------------------------
Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
http://www.cs.waikato.ac.nz/~bernhard                  +64 7 838 4041

_______________________________________________
Wekalist mailing list
Send posts to:
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
)
Hi Bernhard,

Yes you are right, it's a multi-label problem. If I'm not wrong, some
classifiers can handle multi-class problems (more than two class labels)
naturally, and others that are binary class learners need to be called for
each label (ie one-against-the-rest). This can me done using
MultiClassClassifier.

Am I right?

Which classifiers among SMO, NaiveBayes, J48 and kNN can handle multiclass
problems naturally?

2011/7/15 Bernhard Pfahringer <>

> Hi,
>
> there is one slight issue: the categories are NOT mutually exclusive,
> i.e. this is not a multi-class problem, but rather it is a multi-label
> problem,
> or in other words: a document can belong to more than one category.
>
> cheers, Bernhard
>
> On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
> <> wrote:
> > Hi,
> > The reuters-21578 dataset which is available at the weka homepage has
> all
> > the test and train arff files separated by categories. Is it possible to
> > make 1 big file with all the categories? Is there any restriction in
> doing
> > that?
> > Thanks
> > _______________________________________________
> > Wekalist mailing list
> > Send posts to:
> > List info and subscription status:
> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> > List etiquette:
> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> >
> >
>
>
>
> --
> ---------------------------------------------------------------------
> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
> http://www.cs.waikato.ac.nz/~bernhard +64 7 838 4041
>
> _______________________________________________
> Wekalist mailing list
> Send posts to:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>


Hi,

multi-label != multi-class

As you say, most Weka algorithms handle multi-class,
and if not, there is always MultiClassClassifier.

Multi-label is a different story, you'd need to look at
packages like Meka or Mulan, among others:
http://meka.sourceforge.net/
http://mulan.sourceforge.net/

cheers, Bernhard

On Fri, Jul 15, 2011 at 12:19 PM, Diana Iris Calleja
<> wrote:
> Hi Bernhard,
> Yes you are right, it's a multi-label problem. If I'm not wrong, some
> classifiers can handle multi-class problems (more than two class labels)
> naturally, and others that are binary class learners need to be called for
> each label (ie one-against-the-rest). This can me done using
> MultiClassClassifier.
> Am I right?
> Which classifiers among SMO, NaiveBayes, J48 and kNN can handle multiclass
> problems naturally?
>
> 2011/7/15 Bernhard Pfahringer <>
>>
>> Hi,
>>
>> there is one slight issue: the categories are NOT mutually exclusive,
>> i.e. this is not a multi-class problem, but rather it is a multi-label
>> problem,
>> or in other words: a document can belong to more than one category.
>>
>> cheers, Bernhard
>>
>> On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
>> <> wrote:
>> > Hi,
>> > The reuters-21578  dataset which is available at the weka homepage has
>> > all
>> > the test and train arff files separated by categories. Is it possible to
>> > make 1 big file with all the categories? Is there any restriction in
>> > doing
>> > that?
>> > Thanks
>> > _______________________________________________
>> > Wekalist mailing list
>> > Send posts to:
>> > List info and subscription status:
>> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
>> > List etiquette:
>> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>> >
>> >
>>
>>
>>
>> --
>> ---------------------------------------------------------------------
>> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
>> http://www.cs.waikato.ac.nz/~bernhard                  +64 7 838 4041
>>
>> _______________________________________________
>> Wekalist mailing list
>> Send posts to:
>> List info and subscription status:
>> https://list.scms.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:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>
>



--
---------------------------------------------------------------------
Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
http://www.cs.waikato.ac.nz/~bernhard                  +64 7 838 4041

_______________________________________________
Wekalist mailing list
Send posts to:
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
)
Hi Bernhard,

I will try Meka package. I thought I could handle multi-label
with MultiClassClassifier because I read it in this forum:

http://forums.pentaho.com/showthread.php?52105-Overview-of-weka-classifiers-for-a-three-class-problem

So weka ONLY works with binary label cases. Is this right?

Many thanks, Diana.

2011/7/15 Bernhard Pfahringer <>

> Hi,
>
> multi-label != multi-class
>
> As you say, most Weka algorithms handle multi-class,
> and if not, there is always MultiClassClassifier.
>
> Multi-label is a different story, you'd need to look at
> packages like Meka or Mulan, among others:
> http://meka.sourceforge.net/
> http://mulan.sourceforge.net/
>
> cheers, Bernhard
>
> On Fri, Jul 15, 2011 at 12:19 PM, Diana Iris Calleja
> <> wrote:
> > Hi Bernhard,
> > Yes you are right, it's a multi-label problem. If I'm not wrong, some
> > classifiers can handle multi-class problems (more than two class labels)
> > naturally, and others that are binary class learners need to be called
> for
> > each label (ie one-against-the-rest). This can me done using
> > MultiClassClassifier.
> > Am I right?
> > Which classifiers among SMO, NaiveBayes, J48 and kNN can handle
> multiclass
> > problems naturally?
> >
> > 2011/7/15 Bernhard Pfahringer <>
> >>
> >> Hi,
> >>
> >> there is one slight issue: the categories are NOT mutually exclusive,
> >> i.e. this is not a multi-class problem, but rather it is a multi-label
> >> problem,
> >> or in other words: a document can belong to more than one category.
> >>
> >> cheers, Bernhard
> >>
> >> On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
> >> <> wrote:
> >> > Hi,
> >> > The reuters-21578 dataset which is available at the weka homepage has
> >> > all
> >> > the test and train arff files separated by categories. Is it possible
> to
> >> > make 1 big file with all the categories? Is there any restriction in
> >> > doing
> >> > that?
> >> > Thanks
> >> > _______________________________________________
> >> > Wekalist mailing list
> >> > Send posts to:
> >> > List info and subscription status:
> >> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> >> > List etiquette:
> >> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> >> >
> >> >
> >>
> >>
> >>
> >> --
> >> ---------------------------------------------------------------------
> >> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
> >> http://www.cs.waikato.ac.nz/~bernhard +64 7 838 4041
> >>
> >> _______________________________________________
> >> Wekalist mailing list
> >> Send posts to:
> >> List info and subscription status:
> >> https://list.scms.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:
> > List info and subscription status:
> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> > List etiquette:
> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> >
> >
>
>
>
> --
> ---------------------------------------------------------------------
> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
> http://www.cs.waikato.ac.nz/~bernhard +64 7 838 4041
>
> _______________________________________________
> Wekalist mailing list
> Send posts to:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>


On 15/07/11 9:16 PM, Diana Iris Calleja wrote:
> Hi Bernhard,
>
> I will try Meka package. I thought I could handle multi-label
> with MultiClassClassifier because I read it in this forum:
>
> http://forums.pentaho.com/showthread.php?52105-Overview-of-weka-classifiers-for-a-three-class-problem
>
> So weka ONLY works with binary label cases. Is this right?

This is correct. An instance is assumed to belong to only one of a fixed
set of labels.

Cheers,
Mark.

_______________________________________________
Wekalist mailing list
Send posts to:
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
)

Hi,

The reuters-*21578 *dataset which is available at the weka homepage has all
the test and train arff files separated by categories. Is it possible to
make 1 big file with all the categories? Is there any restriction in doing
that?

Thanks


Hi,

there is one slight issue: the categories are NOT mutually exclusive,
i.e. this is not a multi-class problem, but rather it is a multi-label problem,
or in other words: a document can belong to more than one category.

cheers, Bernhard

On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
<> wrote:
> Hi,
> The reuters-21578  dataset which is available at the weka homepage has all
> the test and train arff files separated by categories. Is it possible to
> make 1 big file with all the categories? Is there any restriction in doing
> that?
> Thanks
> _______________________________________________
> Wekalist mailing list
> Send posts to:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>
>



--
---------------------------------------------------------------------
Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
http://www.cs.waikato.ac.nz/~bernhard                  +64 7 838 4041

_______________________________________________
Wekalist mailing list
Send posts to:
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
)
Hi Bernhard,

Yes you are right, it's a multi-label problem. If I'm not wrong, some
classifiers can handle multi-class problems (more than two class labels)
naturally, and others that are binary class learners need to be called for
each label (ie one-against-the-rest). This can me done using
MultiClassClassifier.

Am I right?

Which classifiers among SMO, NaiveBayes, J48 and kNN can handle multiclass
problems naturally?

2011/7/15 Bernhard Pfahringer <>

> Hi,
>
> there is one slight issue: the categories are NOT mutually exclusive,
> i.e. this is not a multi-class problem, but rather it is a multi-label
> problem,
> or in other words: a document can belong to more than one category.
>
> cheers, Bernhard
>
> On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
> <> wrote:
> > Hi,
> > The reuters-21578 dataset which is available at the weka homepage has
> all
> > the test and train arff files separated by categories. Is it possible to
> > make 1 big file with all the categories? Is there any restriction in
> doing
> > that?
> > Thanks
> > _______________________________________________
> > Wekalist mailing list
> > Send posts to:
> > List info and subscription status:
> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> > List etiquette:
> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> >
> >
>
>
>
> --
> ---------------------------------------------------------------------
> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
> http://www.cs.waikato.ac.nz/~bernhard +64 7 838 4041
>
> _______________________________________________
> Wekalist mailing list
> Send posts to:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>


Hi,

multi-label != multi-class

As you say, most Weka algorithms handle multi-class,
and if not, there is always MultiClassClassifier.

Multi-label is a different story, you'd need to look at
packages like Meka or Mulan, among others:
http://meka.sourceforge.net/
http://mulan.sourceforge.net/

cheers, Bernhard

On Fri, Jul 15, 2011 at 12:19 PM, Diana Iris Calleja
<> wrote:
> Hi Bernhard,
> Yes you are right, it's a multi-label problem. If I'm not wrong, some
> classifiers can handle multi-class problems (more than two class labels)
> naturally, and others that are binary class learners need to be called for
> each label (ie one-against-the-rest). This can me done using
> MultiClassClassifier.
> Am I right?
> Which classifiers among SMO, NaiveBayes, J48 and kNN can handle multiclass
> problems naturally?
>
> 2011/7/15 Bernhard Pfahringer <>
>>
>> Hi,
>>
>> there is one slight issue: the categories are NOT mutually exclusive,
>> i.e. this is not a multi-class problem, but rather it is a multi-label
>> problem,
>> or in other words: a document can belong to more than one category.
>>
>> cheers, Bernhard
>>
>> On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
>> <> wrote:
>> > Hi,
>> > The reuters-21578  dataset which is available at the weka homepage has
>> > all
>> > the test and train arff files separated by categories. Is it possible to
>> > make 1 big file with all the categories? Is there any restriction in
>> > doing
>> > that?
>> > Thanks
>> > _______________________________________________
>> > Wekalist mailing list
>> > Send posts to:
>> > List info and subscription status:
>> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
>> > List etiquette:
>> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>> >
>> >
>>
>>
>>
>> --
>> ---------------------------------------------------------------------
>> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
>> http://www.cs.waikato.ac.nz/~bernhard                  +64 7 838 4041
>>
>> _______________________________________________
>> Wekalist mailing list
>> Send posts to:
>> List info and subscription status:
>> https://list.scms.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:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>
>



--
---------------------------------------------------------------------
Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
http://www.cs.waikato.ac.nz/~bernhard                  +64 7 838 4041

_______________________________________________
Wekalist mailing list
Send posts to:
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
)
Hi Bernhard,

I will try Meka package. I thought I could handle multi-label
with MultiClassClassifier because I read it in this forum:

http://forums.pentaho.com/showthread.php?52105-Overview-of-weka-classifiers-for-a-three-class-problem

So weka ONLY works with binary label cases. Is this right?

Many thanks, Diana.

2011/7/15 Bernhard Pfahringer <>

> Hi,
>
> multi-label != multi-class
>
> As you say, most Weka algorithms handle multi-class,
> and if not, there is always MultiClassClassifier.
>
> Multi-label is a different story, you'd need to look at
> packages like Meka or Mulan, among others:
> http://meka.sourceforge.net/
> http://mulan.sourceforge.net/
>
> cheers, Bernhard
>
> On Fri, Jul 15, 2011 at 12:19 PM, Diana Iris Calleja
> <> wrote:
> > Hi Bernhard,
> > Yes you are right, it's a multi-label problem. If I'm not wrong, some
> > classifiers can handle multi-class problems (more than two class labels)
> > naturally, and others that are binary class learners need to be called
> for
> > each label (ie one-against-the-rest). This can me done using
> > MultiClassClassifier.
> > Am I right?
> > Which classifiers among SMO, NaiveBayes, J48 and kNN can handle
> multiclass
> > problems naturally?
> >
> > 2011/7/15 Bernhard Pfahringer <>
> >>
> >> Hi,
> >>
> >> there is one slight issue: the categories are NOT mutually exclusive,
> >> i.e. this is not a multi-class problem, but rather it is a multi-label
> >> problem,
> >> or in other words: a document can belong to more than one category.
> >>
> >> cheers, Bernhard
> >>
> >> On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
> >> <> wrote:
> >> > Hi,
> >> > The reuters-21578 dataset which is available at the weka homepage has
> >> > all
> >> > the test and train arff files separated by categories. Is it possible
> to
> >> > make 1 big file with all the categories? Is there any restriction in
> >> > doing
> >> > that?
> >> > Thanks
> >> > _______________________________________________
> >> > Wekalist mailing list
> >> > Send posts to:
> >> > List info and subscription status:
> >> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> >> > List etiquette:
> >> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> >> >
> >> >
> >>
> >>
> >>
> >> --
> >> ---------------------------------------------------------------------
> >> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
> >> http://www.cs.waikato.ac.nz/~bernhard +64 7 838 4041
> >>
> >> _______________________________________________
> >> Wekalist mailing list
> >> Send posts to:
> >> List info and subscription status:
> >> https://list.scms.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:
> > List info and subscription status:
> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> > List etiquette:
> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> >
> >
>
>
>
> --
> ---------------------------------------------------------------------
> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
> http://www.cs.waikato.ac.nz/~bernhard +64 7 838 4041
>
> _______________________________________________
> Wekalist mailing list
> Send posts to:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>


On 15/07/11 9:16 PM, Diana Iris Calleja wrote:
> Hi Bernhard,
>
> I will try Meka package. I thought I could handle multi-label
> with MultiClassClassifier because I read it in this forum:
>
> http://forums.pentaho.com/showthread.php?52105-Overview-of-weka-classifiers-for-a-three-class-problem
>
> So weka ONLY works with binary label cases. Is this right?

This is correct. An instance is assumed to belong to only one of a fixed
set of labels.

Cheers,
Mark.

_______________________________________________
Wekalist mailing list
Send posts to:
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
)
Hi Mark/Bernhard,

Strong apologies for asking again, but I'm a bit confused.

What I need to do is to classify the instances in one of a fixed set of
labels, but I have 83 labels. Therefore, in the training set I will have
instances with only one label among those 83, and the same in the testing
set.

My class attribute in my arff file for the reuters collection would
be: @attribute class
{acq,earn,gas,gold,interest,trade,ship,coffee,jet,silver,crude,money-fx,nat-gas,wheat,carcass,cocoa,strategic-metal,sorghum,corn,sugar,tin,soybean,grain,meal-feed,rubber,livestock,veg-oil,gnp,alum,iron-steel,groundnut,money-supply,ipi,fuel,pet-chem,barley,reserves,yen,copper,jobs,rice,dlr,rapeseed,platinum,wpi,cpi,palm-oil,bop,cotton,rape-oil,oat,nzdlr,sunseed,retail,l-cattle,nickel,dmk,lead,oilseed,orange,housing,soy-meal,zinc,lumber,coconut,groundnut-oil,tea,propane,lei,instal-debt,heat,income,soy-oil,hog,copra-cake,rand,nkr,coconut-oil,naphtha,potato,castor-oil,cpu,palladium,sun-oil,cotton-oil,dfl}

Can I handle this with Weka directly (with multi class classifier or other)
or I need a third party project like mulan o meka ?

Many thanks, Diana

2011/7/15 Mark Hall <>

> On 15/07/11 9:16 PM, Diana Iris Calleja wrote:
>
>> Hi Bernhard,
>>
>> I will try Meka package. I thought I could handle multi-label
>> with MultiClassClassifier because I read it in this forum:
>>
>> http://forums.pentaho.com/**showthread.php?52105-Overview-**
>> of-weka-classifiers-for-a-**three-class-problem
>>
>> So weka ONLY works with binary label cases. Is this right?
>>
>
> This is correct. An instance is assumed to belong to only one of a fixed
> set of labels.
>
> Cheers,
> Mark.
>
>
> ______________________________**_________________
> Wekalist mailing list
> Send posts to: .**nz<>
> List info and subscription status: https://list.scms.waikato.ac.**
> nz/mailman/listinfo/wekalist
> List etiquette: http://www.cs.waikato.ac.nz/~**
> ml/weka/mailinglist_etiquette.**html
>

Hi,

The reuters-*21578 *dataset which is available at the weka homepage has all
the test and train arff files separated by categories. Is it possible to
make 1 big file with all the categories? Is there any restriction in doing
that?

Thanks


Hi,

there is one slight issue: the categories are NOT mutually exclusive,
i.e. this is not a multi-class problem, but rather it is a multi-label problem,
or in other words: a document can belong to more than one category.

cheers, Bernhard

On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
<> wrote:
> Hi,
> The reuters-21578  dataset which is available at the weka homepage has all
> the test and train arff files separated by categories. Is it possible to
> make 1 big file with all the categories? Is there any restriction in doing
> that?
> Thanks
> _______________________________________________
> Wekalist mailing list
> Send posts to:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>
>



--
---------------------------------------------------------------------
Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
http://www.cs.waikato.ac.nz/~bernhard                  +64 7 838 4041

_______________________________________________
Wekalist mailing list
Send posts to:
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
)
Hi Bernhard,

Yes you are right, it's a multi-label problem. If I'm not wrong, some
classifiers can handle multi-class problems (more than two class labels)
naturally, and others that are binary class learners need to be called for
each label (ie one-against-the-rest). This can me done using
MultiClassClassifier.

Am I right?

Which classifiers among SMO, NaiveBayes, J48 and kNN can handle multiclass
problems naturally?

2011/7/15 Bernhard Pfahringer <>

> Hi,
>
> there is one slight issue: the categories are NOT mutually exclusive,
> i.e. this is not a multi-class problem, but rather it is a multi-label
> problem,
> or in other words: a document can belong to more than one category.
>
> cheers, Bernhard
>
> On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
> <> wrote:
> > Hi,
> > The reuters-21578 dataset which is available at the weka homepage has
> all
> > the test and train arff files separated by categories. Is it possible to
> > make 1 big file with all the categories? Is there any restriction in
> doing
> > that?
> > Thanks
> > _______________________________________________
> > Wekalist mailing list
> > Send posts to:
> > List info and subscription status:
> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> > List etiquette:
> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> >
> >
>
>
>
> --
> ---------------------------------------------------------------------
> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
> http://www.cs.waikato.ac.nz/~bernhard +64 7 838 4041
>
> _______________________________________________
> Wekalist mailing list
> Send posts to:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>


Hi,

multi-label != multi-class

As you say, most Weka algorithms handle multi-class,
and if not, there is always MultiClassClassifier.

Multi-label is a different story, you'd need to look at
packages like Meka or Mulan, among others:
http://meka.sourceforge.net/
http://mulan.sourceforge.net/

cheers, Bernhard

On Fri, Jul 15, 2011 at 12:19 PM, Diana Iris Calleja
<> wrote:
> Hi Bernhard,
> Yes you are right, it's a multi-label problem. If I'm not wrong, some
> classifiers can handle multi-class problems (more than two class labels)
> naturally, and others that are binary class learners need to be called for
> each label (ie one-against-the-rest). This can me done using
> MultiClassClassifier.
> Am I right?
> Which classifiers among SMO, NaiveBayes, J48 and kNN can handle multiclass
> problems naturally?
>
> 2011/7/15 Bernhard Pfahringer <>
>>
>> Hi,
>>
>> there is one slight issue: the categories are NOT mutually exclusive,
>> i.e. this is not a multi-class problem, but rather it is a multi-label
>> problem,
>> or in other words: a document can belong to more than one category.
>>
>> cheers, Bernhard
>>
>> On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
>> <> wrote:
>> > Hi,
>> > The reuters-21578  dataset which is available at the weka homepage has
>> > all
>> > the test and train arff files separated by categories. Is it possible to
>> > make 1 big file with all the categories? Is there any restriction in
>> > doing
>> > that?
>> > Thanks
>> > _______________________________________________
>> > Wekalist mailing list
>> > Send posts to:
>> > List info and subscription status:
>> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
>> > List etiquette:
>> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>> >
>> >
>>
>>
>>
>> --
>> ---------------------------------------------------------------------
>> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
>> http://www.cs.waikato.ac.nz/~bernhard                  +64 7 838 4041
>>
>> _______________________________________________
>> Wekalist mailing list
>> Send posts to:
>> List info and subscription status:
>> https://list.scms.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:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>
>



--
---------------------------------------------------------------------
Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
http://www.cs.waikato.ac.nz/~bernhard                  +64 7 838 4041

_______________________________________________
Wekalist mailing list
Send posts to:
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
)
Hi Bernhard,

I will try Meka package. I thought I could handle multi-label
with MultiClassClassifier because I read it in this forum:

http://forums.pentaho.com/showthread.php?52105-Overview-of-weka-classifiers-for-a-three-class-problem

So weka ONLY works with binary label cases. Is this right?

Many thanks, Diana.

2011/7/15 Bernhard Pfahringer <>

> Hi,
>
> multi-label != multi-class
>
> As you say, most Weka algorithms handle multi-class,
> and if not, there is always MultiClassClassifier.
>
> Multi-label is a different story, you'd need to look at
> packages like Meka or Mulan, among others:
> http://meka.sourceforge.net/
> http://mulan.sourceforge.net/
>
> cheers, Bernhard
>
> On Fri, Jul 15, 2011 at 12:19 PM, Diana Iris Calleja
> <> wrote:
> > Hi Bernhard,
> > Yes you are right, it's a multi-label problem. If I'm not wrong, some
> > classifiers can handle multi-class problems (more than two class labels)
> > naturally, and others that are binary class learners need to be called
> for
> > each label (ie one-against-the-rest). This can me done using
> > MultiClassClassifier.
> > Am I right?
> > Which classifiers among SMO, NaiveBayes, J48 and kNN can handle
> multiclass
> > problems naturally?
> >
> > 2011/7/15 Bernhard Pfahringer <>
> >>
> >> Hi,
> >>
> >> there is one slight issue: the categories are NOT mutually exclusive,
> >> i.e. this is not a multi-class problem, but rather it is a multi-label
> >> problem,
> >> or in other words: a document can belong to more than one category.
> >>
> >> cheers, Bernhard
> >>
> >> On Fri, Jul 15, 2011 at 10:11 AM, Diana Iris Calleja
> >> <> wrote:
> >> > Hi,
> >> > The reuters-21578 dataset which is available at the weka homepage has
> >> > all
> >> > the test and train arff files separated by categories. Is it possible
> to
> >> > make 1 big file with all the categories? Is there any restriction in
> >> > doing
> >> > that?
> >> > Thanks
> >> > _______________________________________________
> >> > Wekalist mailing list
> >> > Send posts to:
> >> > List info and subscription status:
> >> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> >> > List etiquette:
> >> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> >> >
> >> >
> >>
> >>
> >>
> >> --
> >> ---------------------------------------------------------------------
> >> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
> >> http://www.cs.waikato.ac.nz/~bernhard +64 7 838 4041
> >>
> >> _______________________________________________
> >> Wekalist mailing list
> >> Send posts to:
> >> List info and subscription status:
> >> https://list.scms.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:
> > List info and subscription status:
> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> > List etiquette:
> > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> >
> >
>
>
>
> --
> ---------------------------------------------------------------------
> Bernhard Pfahringer, Dept. of Computer Science, University of Waikato
> http://www.cs.waikato.ac.nz/~bernhard +64 7 838 4041
>
> _______________________________________________
> Wekalist mailing list
> Send posts to:
> List info and subscription status:
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>


On 15/07/11 9:16 PM, Diana Iris Calleja wrote:
> Hi Bernhard,
>
> I will try Meka package. I thought I could handle multi-label
> with MultiClassClassifier because I read it in this forum:
>
> http://forums.pentaho.com/showthread.php?52105-Overview-of-weka-classifiers-for-a-three-class-problem
>
> So weka ONLY works with binary label cases. Is this right?

This is correct. An instance is assumed to belong to only one of a fixed
set of labels.

Cheers,
Mark.

_______________________________________________
Wekalist mailing list
Send posts to:
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
)
Hi Mark/Bernhard,

Strong apologies for asking again, but I'm a bit confused.

What I need to do is to classify the instances in one of a fixed set of
labels, but I have 83 labels. Therefore, in the training set I will have
instances with only one label among those 83, and the same in the testing
set.

My class attribute in my arff file for the reuters collection would
be: @attribute class
{acq,earn,gas,gold,interest,trade,ship,coffee,jet,silver,crude,money-fx,nat-gas,wheat,carcass,cocoa,strategic-metal,sorghum,corn,sugar,tin,soybean,grain,meal-feed,rubber,livestock,veg-oil,gnp,alum,iron-steel,groundnut,money-supply,ipi,fuel,pet-chem,barley,reserves,yen,copper,jobs,rice,dlr,rapeseed,platinum,wpi,cpi,palm-oil,bop,cotton,rape-oil,oat,nzdlr,sunseed,retail,l-cattle,nickel,dmk,lead,oilseed,orange,housing,soy-meal,zinc,lumber,coconut,groundnut-oil,tea,propane,lei,instal-debt,heat,income,soy-oil,hog,copra-cake,rand,nkr,coconut-oil,naphtha,potato,castor-oil,cpu,palladium,sun-oil,cotton-oil,dfl}

Can I handle this with Weka directly (with multi class classifier or other)
or I need a third party project like mulan o meka ?

Many thanks, Diana

2011/7/15 Mark Hall <>

> On 15/07/11 9:16 PM, Diana Iris Calleja wrote:
>
>> Hi Bernhard,
>>
>> I will try Meka package. I thought I could handle multi-label
>> with MultiClassClassifier because I read it in this forum:
>>
>> http://forums.pentaho.com/**showthread.php?52105-Overview-**
>> of-weka-classifiers-for-a-**three-class-problem
>>
>> So weka ONLY works with binary label cases. Is this right?
>>
>
> This is correct. An instance is assumed to belong to only one of a fixed
> set of labels.
>
> Cheers,
> Mark.
>
>
> ______________________________**_________________
> Wekalist mailing list
> Send posts to: .**nz<>
> List info and subscription status: https://list.scms.waikato.ac.**
> nz/mailman/listinfo/wekalist
> List etiquette: http://www.cs.waikato.ac.nz/~**
> ml/weka/mailinglist_etiquette.**html
>


On 15/07/11 10:10 PM, Diana Iris Calleja wrote:
> Hi Mark/Bernhard,
>
> Strong apologies for asking again, but I'm a bit confused.
>
> What I need to do is to classify the instances in one of a fixed set of
> labels, but I have 83 labels. Therefore, in the training set I will have
> instances with only one label among those 83, and the same in the
> testing set.
>
> My class attribute in my arff file for the reuters collection would
> be: @attribute class
> {acq,earn,gas,gold,interest,trade,ship,coffee,jet,silver,crude,money-fx,nat-gas,wheat,carcass,cocoa,strategic-metal,sorghum,corn,sugar,tin,soybean,grain,meal-feed,rubber,livestock,veg-oil,gnp,alum,iron-steel,groundnut,money-supply,ipi,fuel,pet-chem,barley,reserves,yen,copper,jobs,rice,dlr,rapeseed,platinum,wpi,cpi,palm-oil,bop,cotton,rape-oil,oat,nzdlr,sunseed,retail,l-cattle,nickel,dmk,lead,oilseed,orange,housing,soy-meal,zinc,lumber,coconut,groundnut-oil,tea,propane,lei,instal-debt,heat,income,soy-oil,hog,copra-cake,rand,nkr,coconut-oil,naphtha,potato,castor-oil,cpu,palladium,sun-oil,cotton-oil,dfl}
>
> Can I handle this with Weka directly (with multi class classifier or
> other) or I need a third party project like mulan o meka ?

Weka can handle this situation. The issue with the reuters-21578 data,
as Bernhard mentioned to begin with, is that the classes are not
mutually exclusive. This means that the the same instance may be in more
than one of the train/test files (since it belongs to more than one
category).

Cheers,
Mark.

_______________________________________________
Wekalist mailing list
Send posts to:
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
)