Cluster tab

classic Classic list List threaded Threaded
7 messages Options
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Cluster tab

Edward Wiskers

Hi all,

In need to ask about a specific functionality in Weka's cluster tab.

Since a cluster algorithm doesn't process the class (based on the definition  of unsupervised learning), suppose even when loading a data with class that has labels, any cluster method should ignore this class by default.

My question, if the upper issue is applied in Weka's cluster approach, then why in the "Ignore attributes" option we can still see the class attribute gets mentioned with other number of attribute?

And why if this class is selected to be ignored and then it's ignored that will produce another result (obtained before ignoring the class)?

Any help would be highly appreciated.

Edward


_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Cluster tab

Eibe Frank-2
Administrator
The class setting does not get transferred between Explorer panels.

The “Preprocess", “Classify", "Select attributes” and “Visualize” all treat the last attribute as the class by default. This needs to be changed in each tab individually.

By default, the “Cluster” tab assumes that none of the attributes is the class. This can be changed by ticking “Classes to clusters evaluation”.

The “Associate” tab assumes that there are no classes in the data.

Cheers,
Eibe

> On 3/08/2017, at 11:31 AM, Edward Wiskers <[hidden email]> wrote:
>
> Hi all,
>
> In need to ask about a specific functionality in Weka's cluster tab.
>
> Since a cluster algorithm doesn't process the class (based on the definition  of unsupervised learning), suppose even when loading a data with class that has labels, any cluster method should ignore this class by default.
>
> My question, if the upper issue is applied in Weka's cluster approach, then why in the "Ignore attributes" option we can still see the class attribute gets mentioned with other number of attribute?
>
> And why if this class is selected to be ignored and then it's ignored that will produce another result (obtained before ignoring the class)?
>
> Any help would be highly appreciated.
>
> Edward
>
> _______________________________________________
> Wekalist mailing list
> Send posts to: [hidden email]
> List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html

_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Cluster tab

Edward Wiskers

Thanks you very much, Eibe, for the prompt reply.

> By default, the “Cluster” tab assumes that none of the attributes is the class. This can be changed by ticking “Classes to clusters evaluation”.
>

Do you mean that, a clusterer method will get all the attribute without any assumption about the last attitude as a class?

> The “Associate” tab assumes that there are no classes in the data.
>
Do you mean here that the association rules method will automatically makes an assumption about the data through considering the last attribute is a class? And based on that the algorithm ignores the last attribute automatically?

Why such difference between cluster and association rules tab since both are unsupervised learning techniques where both behave similarly in terms of ignoring the class (last attribute)?

Edward

> Cheers,
> Eibe
>
> > On 3/08/2017, at 11:31 AM, Edward Wiskers <[hidden email]> wrote:
> >
> > Hi all,
> >
> > In need to ask about a specific functionality in Weka's cluster tab.
> >
> > Since a cluster algorithm doesn't process the class (based on the definition  of unsupervised learning), suppose even when loading a data with class that has labels, any cluster method should ignore this class by default.
> >
> > My question, if the upper issue is applied in Weka's cluster approach, then why in the "Ignore attributes" option we can still see the class attribute gets mentioned with other number of attribute?
> >
> > And why if this class is selected to be ignored and then it's ignored that will produce another result (obtained before ignoring the class)?
> >
> > Any help would be highly appreciated.
> >
> > Edward
> >
> > _______________________________________________
> > Wekalist mailing list
> > Send posts to: [hidden email]
> > List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
> > List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>
> _______________________________________________
> Wekalist mailing list
> Send posts to: [hidden email]
> List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html


_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Cluster tab

Eibe Frank-2
Administrator

> On 3/08/2017, at 12:50 PM, Edward Wiskers <[hidden email]> wrote:
>
> > By default, the “Cluster” tab assumes that none of the attributes is the class. This can be changed by ticking “Classes to clusters evaluation”.
> >
>
> Do you mean that, a clusterer method will get all the attribute without any assumption about the last attitude as a class?

Correct, it assumes that there is no class attribute (by default). All attributes in the dataset are treated equally.

> > The “Associate” tab assumes that there are no classes in the data.
> >
> Do you mean here that the association rules method will automatically makes an assumption about the data through considering the last attribute is a class? And based on that the algorithm ignores the last attribute automatically?
>
> Why such difference between cluster and association rules tab since both are unsupervised learning techniques where both behave similarly in terms of ignoring the class (last attribute)?

There is no difference (by default). Both assume that there is no class attribute.

Cheers,
Eibe

_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Cluster tab

Edward Wiskers


> There is no difference (by default). Both assume that there is no class attribute.

So why we need to ignore the class in the Cluster tab but not in the Associate tab?

Edward
>
> Cheers,
> Eibe
>
> _______________________________________________
> Wekalist mailing list
> Send posts to: [hidden email]
> List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html


_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Cluster tab

Eibe Frank-2
Administrator
You are right, I suppose it would be nice to have the option to ignore attributes in the Associate tab as well.

Cheers,
Eibe

> On 3/08/2017, at 1:08 PM, Edward Wiskers <[hidden email]> wrote:
>
>
> > There is no difference (by default). Both assume that there is no class attribute.
>
> So why we need to ignore the class in the Cluster tab but not in the Associate tab?
>
> Edward
> >
> > Cheers,
> > Eibe
> >
> > _______________________________________________
> > Wekalist mailing list
> > Send posts to: [hidden email]
> > List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
> > List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> _______________________________________________
> Wekalist mailing list
> Send posts to: [hidden email]
> List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html

_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Cluster tab

Edward Wiskers

That's right, Eibe. This exactly what I meant, in order to make the behaviour of unsupervised learning extremely clear for the user when working in the Cluster and the Associate tabs.

Hope this issue will be considered in soon Weka versions.

Kind regards,
Edward

On 3 Aug 2017 9:13 a.m., "Eibe Frank" <[hidden email]> wrote:
You are right, I suppose it would be nice to have the option to ignore attributes in the Associate tab as well.

Cheers,
Eibe

> On 3/08/2017, at 1:08 PM, Edward Wiskers <[hidden email]> wrote:
>
>
> > There is no difference (by default). Both assume that there is no class attribute.
>
> So why we need to ignore the class in the Cluster tab but not in the Associate tab?
>
> Edward
> >
> > Cheers,
> > Eibe
> >
> > _______________________________________________
> > Wekalist mailing list
> > Send posts to: [hidden email]
> > List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
> > List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> _______________________________________________
> Wekalist mailing list
> Send posts to: [hidden email]
> List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html

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

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