Help for attribute discretize in knowledge flow

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

Help for attribute discretize in knowledge flow

Stanley, Yemin Shi
Hi,
I am using weka knowledge flow for attribute discretization, but it seems to stop at this procedure: the knowledge flowchat is like this:
 
Arff loader ----(dataset)---->
(feature) discretize ----(dataset)---->
class assigner ----(dataset)---->
crossvalidation fold maker ----(training set/ testing set)---->
j48 classifier ----(batch classifier)---->
classifier performance evlauation ----(text)---->
 text viewer.
 
I only changed the discretize procedure attributeIndices parameter from first-last to: first-34, the dataset ionosphere.arff is downloaded from the weka website.
 
Could anybody help me on this,
 
BTW, if I remove the (feature) discretize procedure, everything works.
 
Arff loader ----(dataset)---->
class assigner ----(dataset)---->
 crossvalidation fold maker ----(training set/ testing set)---->
j48 classifier ----(batch classifier)---->
classifier performance evlauation ----(text)---->
text viewer.
 
Thanks for time
Yemin


Sell on Yahoo! Auctions - No fees. Bid on great items.
_______________________________________________
Wekalist mailing list
[hidden email]
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
Reply | Threaded
Open this post in threaded view
|

Re: Help for attribute discretize in knowledge flow

Mark Hall-11
Which discretize filter are you using? If you are using the one in
weka.filters.supervised then the behaviour (i.e not working) is
correct as there is no class attribute defined in the dataset at this
point. There should be a message printed to the log (this isn't the
case and needs to be fixed).

You can use the discretize filter in weka.filters.unsupervised and it
will work correctly, or you can put a class assigner right after the
arff loader in order to use the supervised version. However, if you do
the latter (i.e. use the supervised discretize) and you do a cross
validation later in the flow then you are effectively cheating and
your results will be overly optimistic. This is because the supervised
filter will have learned from the test data.

Cheers,
Mark.

On 7/11/05, Stanley, Yemin Shi <[hidden email]> wrote:

> Hi,
> I am using weka knowledge flow for attribute discretization, but it seems to
> stop at this procedure: the knowledge flowchat is like this:
>  
> Arff loader ----(dataset)---->
> (feature) discretize ----(dataset)---->
> class assigner ----(dataset)---->
> crossvalidation fold maker ----(training set/ testing set)---->
> j48 classifier ----(batch classifier)---->
> classifier performance evlauation ----(text)---->
>  text viewer.
>  
> I only changed the discretize procedure attributeIndices parameter from
> first-last to: first-34, the dataset ionosphere.arff is downloaded from the
> weka website.
>  
> Could anybody help me on this,
>  
> BTW, if I remove the (feature) discretize procedure, everything works.
>  
>  
> Arff loader ----(dataset)---->
> class assigner ----(dataset)---->
>  crossvalidation fold maker ----(training set/ testing set)---->
> j48 classifier ----(batch classifier)---->
> classifier performance evlauation ----(text)---->
> text viewer.
>  
> Thanks for time
> Yemin
>
>  ________________________________
>  Sell on Yahoo! Auctions - No fees. Bid on great items.
>
>
> _______________________________________________
> Wekalist mailing list
> [hidden email]
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
>
>
>


--
Mark Hall
Department of Computer Science
University of Waikato
Hamilton
New Zealand
www.cs.waikato.ac.nz

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