Weka 101

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

Weka 101

Peppy & Siuly Adi-Purnomo
Hi there

I am completing my Masters in Stats and Weka is one software that I am
investigating when comes to analysing Classification problem.

Is there anything out there for me to read on how to use Weka?  I know that
there is a book that is sold in Amazon etc but at this stage I do not have
any budget to invest in that.  I just want to explore at the skin level
first....

Much appreciate for any inputs.

Peppy



_______________________________________________
Wekalist mailing list
[hidden email]
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
Reply | Threaded
Open this post in threaded view
|

Re: Weka 101

Paul m
I would suggest you download weka, have a look at the user manuals
that come with the download and then read the source code. This will
give you an indepth insight into how it works. The user manuals will
provide the 'skin level' overview.

Have fun.. /paul

On 7/5/05, Peppy & Siuly Adi-Purnomo <[hidden email]> wrote:

> Hi there
>
> I am completing my Masters in Stats and Weka is one software that I am
> investigating when comes to analysing Classification problem.
>
> Is there anything out there for me to read on how to use Weka?  I know that
> there is a book that is sold in Amazon etc but at this stage I do not have
> any budget to invest in that.  I just want to explore at the skin level
> first....
>
> Much appreciate for any inputs.
>
> Peppy
>
>
>
> _______________________________________________
> Wekalist mailing list
> [hidden email]
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
>

_______________________________________________
Wekalist mailing list
[hidden email]
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
Reply | Threaded
Open this post in threaded view
|

Re: Weka 101

JRijnberk
In reply to this post by Peppy & Siuly Adi-Purnomo
Hi

Rather start using the GUI version (download and install)and get the book
from the library. Obtain a birds eye view from it and decide what to do next
from there.

Hans van Rijnberk
Assort Vision

At 02:22 PM 06/07/2005 +1200, Paul wrote:

>I would suggest you download weka, have a look at the user manuals
>that come with the download and then read the source code. This will
>give you an indepth insight into how it works. The user manuals will
>provide the 'skin level' overview.
>
>Have fun.. /paul
>
>On 7/5/05, Peppy & Siuly Adi-Purnomo <[hidden email]> wrote:
>> Hi there
>>
>> I am completing my Masters in Stats and Weka is one software that I am
>> investigating when comes to analysing Classification problem.
>>
>> Is there anything out there for me to read on how to use Weka?  I know that
>> there is a book that is sold in Amazon etc but at this stage I do not have
>> any budget to invest in that.  I just want to explore at the skin level
>> first....
>>
>> Much appreciate for any inputs.
>>
>> Peppy
>>
>>
>>
>> _______________________________________________
>> Wekalist mailing list
>> [hidden email]
>> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
>>
>
>_______________________________________________
>Wekalist mailing list
>[hidden email]
>https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
>
>


Hans van Rijnberk

[hidden email]



_______________________________________________
Wekalist mailing list
[hidden email]
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
Reply | Threaded
Open this post in threaded view
|

J48 and unseen cases

rich@thevillas.eclipse.co.uk
In reply to this post by Paul m
Hi, i'm pretty sure this isn't possible, maybe someone could confirm? I
want to make predictions to unseen cases using J48. So, I have my test
and training set that I perform 10 fold x validation on and then i'd
like to apply this to cases where the class is unknown. I know this is
possible with C5, but not neccessarily J48 as this was built on c4.5
revision 8. Is this correct?

Is there another classifier that I can use?

thanks in advance,

Rich

_______________________________________________
Wekalist mailing list
[hidden email]
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
Reply | Threaded
Open this post in threaded view
|

Re: J48 and unseen cases

Santi Planet
I think that it's possible to cassify instances whose class is unknown with
the J48 classifier and with any other classifier. You only have to set the
test set what you want to evaluate in the Weka Explorer.

I wish this will help you.

Santi.

-----Mensaje original-----
De: [hidden email]
[mailto:[hidden email]] En nombre de rich
Enviado el: miƩrcoles, 06 de julio de 2005 17:14
Para: [hidden email]
Asunto: [Wekalist] J48 and unseen cases

Hi, i'm pretty sure this isn't possible, maybe someone could confirm? I
want to make predictions to unseen cases using J48. So, I have my test
and training set that I perform 10 fold x validation on and then i'd
like to apply this to cases where the class is unknown. I know this is
possible with C5, but not neccessarily J48 as this was built on c4.5
revision 8. Is this correct?

Is there another classifier that I can use?

thanks in advance,

Rich

_______________________________________________
Wekalist mailing list
[hidden email]
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist



_______________________________________________
Wekalist mailing list
[hidden email]
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
Reply | Threaded
Open this post in threaded view
|

classifiers learning only from one type of instances

Grazia Russo-Lassner

Hi,

are there classifiers that perform well even in the presence of a class
imbalance problem ?

Trying to solve the class imbalance problem (by rebalancing in some way
the class that is overpresent) seems at times on certain applications kind
of ad hoc and not really satisfactory in the end results.
So, if there is such a classifier, could you please send me a pointer to
it ? And if there isn;t one, could you please let me know why  (still a
novice of machine learning) ?

Thank you,

Grazia



_______________________________________________
Wekalist mailing list
[hidden email]
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
Reply | Threaded
Open this post in threaded view
|

Re: classifiers learning only from one type of instances

Paul m
The topic of class imbalance in data sets is actually the basis of my
masters thesis.

There is no one classifier that performs well. Often research has
suggested that utilising some type of sampling technique (ie, over
sampling the minority class or undersampling the majority class)
combined with a base classifer is the best approach.

There is however plenty of work going on in this area....

/paul

On 7/7/05, Grazia Russo-Lassner <[hidden email]> wrote:

>
> Hi,
>
> are there classifiers that perform well even in the presence of a class
> imbalance problem ?
>
> Trying to solve the class imbalance problem (by rebalancing in some way
> the class that is overpresent) seems at times on certain applications kind
> of ad hoc and not really satisfactory in the end results.
> So, if there is such a classifier, could you please send me a pointer to
> it ? And if there isn;t one, could you please let me know why  (still a
> novice of machine learning) ?
>
> Thank you,
>
> Grazia
>
>
>
> _______________________________________________
> Wekalist mailing list
> [hidden email]
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
>

_______________________________________________
Wekalist mailing list
[hidden email]
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
Reply | Threaded
Open this post in threaded view
|

Rename attribute

Clifford Lyon
In reply to this post by Santi Planet
What's the standard way to (programatically) rename an attribute?

Looks like you'd need to copy the attribute with a new name, and drop
the old one?

tia


_______________________________________________
Wekalist mailing list
[hidden email]
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
Reply | Threaded
Open this post in threaded view
|

Re: Rename attribute

Peter Reutemann
> What's the standard way to (programatically) rename an attribute?

Check out the renamAttribute(...) methods of the class weka.core.Instances:
    http://weka.sourceforge.net/doc/weka/core/Instances.html

HTH

Cheers, Peter

--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/     +64 (7) 838-4466 Ext. 5174

_______________________________________________
Wekalist mailing list
[hidden email]
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
Reply | Threaded
Open this post in threaded view
|

Re: classifiers learning only from one type of instances

Grazia Russo-Lassner
In reply to this post by Paul m

what about the one-class classification ?
See for instance, http://www.csie.ntu.edu.tw/~cjlin/libsvm/

Grazia


On Thu, 7 Jul 2005, Paul wrote:

> The topic of class imbalance in data sets is actually the basis of my
> masters thesis.
>
> There is no one classifier that performs well. Often research has
> suggested that utilising some type of sampling technique (ie, over
> sampling the minority class or undersampling the majority class)
> combined with a base classifer is the best approach.
>
> There is however plenty of work going on in this area....
>
> /paul
>
> On 7/7/05, Grazia Russo-Lassner <[hidden email]> wrote:
>>
>> Hi,
>>
>> are there classifiers that perform well even in the presence of a class
>> imbalance problem ?
>>
>> Trying to solve the class imbalance problem (by rebalancing in some way
>> the class that is overpresent) seems at times on certain applications kind
>> of ad hoc and not really satisfactory in the end results.
>> So, if there is such a classifier, could you please send me a pointer to
>> it ? And if there isn;t one, could you please let me know why  (still a
>> novice of machine learning) ?
>>
>> Thank you,
>>
>> Grazia
>>
>>
>>
>> _______________________________________________
>> Wekalist mailing list
>> [hidden email]
>> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
>>
>

_______________________________________________
Wekalist mailing list
[hidden email]
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
Reply | Threaded
Open this post in threaded view
|

Re: Rename attribute

Clifford Lyon
In reply to this post by Peter Reutemann
ah, thanks!!  (I had been looking at the weka.core.Attribute class)

Peter Reutemann wrote:

>> What's the standard way to (programatically) rename an attribute?
>
>
> Check out the renamAttribute(...) methods of the class weka.core.Instances:
>    http://weka.sourceforge.net/doc/weka/core/Instances.html
>
> HTH
>
> Cheers, Peter
>

_______________________________________________
Wekalist mailing list
[hidden email]
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
Reply | Threaded
Open this post in threaded view
|

Re: classifiers learning only from one type of instances

Paul m
In reply to this post by Grazia Russo-Lassner
Imbalanced data sets are typical 2 or more classes. One class training
(whether you train the minority or majority class) will always
comprise the model when identifying new, previously unseen, instances.

There has been previous research about only training one of the
classes, and although provided suitable results, nothing spectacular.
I had a look at :
http://www.kernel-machines.org/jmlr/volume2/manevitz01a/manevitz01a.pdf
 and got the impression that typically one-class problems are where
more classes are unavailable as opposed to the preferred manner to
deal with class imbalance.

/paul

On 7/7/05, Grazia Russo-Lassner <[hidden email]> wrote:

>
> what about the one-class classification ?
> See for instance, http://www.csie.ntu.edu.tw/~cjlin/libsvm/
>
> Grazia
>
>
> On Thu, 7 Jul 2005, Paul wrote:
>
> > The topic of class imbalance in data sets is actually the basis of my
> > masters thesis.
> >
> > There is no one classifier that performs well. Often research has
> > suggested that utilising some type of sampling technique (ie, over
> > sampling the minority class or undersampling the majority class)
> > combined with a base classifer is the best approach.
> >
> > There is however plenty of work going on in this area....
> >
> > /paul
> >
> > On 7/7/05, Grazia Russo-Lassner <[hidden email]> wrote:
> >>
> >> Hi,
> >>
> >> are there classifiers that perform well even in the presence of a class
> >> imbalance problem ?
> >>
> >> Trying to solve the class imbalance problem (by rebalancing in some way
> >> the class that is overpresent) seems at times on certain applications kind
> >> of ad hoc and not really satisfactory in the end results.
> >> So, if there is such a classifier, could you please send me a pointer to
> >> it ? And if there isn;t one, could you please let me know why  (still a
> >> novice of machine learning) ?
> >>
> >> Thank you,
> >>
> >> Grazia
> >>
> >>
> >>
> >> _______________________________________________
> >> Wekalist mailing list
> >> [hidden email]
> >> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> >>
> >
>

_______________________________________________
Wekalist mailing list
[hidden email]
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist