Re: classifier for working out which combinations ofattributes give lowest error rate

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

Re: classifier for working out which combinations ofattributes give lowest error rate

JRijnberk


At 06:11 PM 21/07/2005 +0530, Nanda, Subrat (Research) wrote:
>Hi Rich,
>
>A couple of approaches consistent with what you are asking are the Forward
and Backward Feature selection procedures. Briefly, these approaches start
with either a null set (Forward) or complete (Backward) set of selected
attributes and then add or delete feature sets respectively and incrementally.
>
>These methods usually take into account the learning method to be used, and
so are 'wrapper' based methods in the sense that they do take the learning
algorithm into feature ranking. A different versions of this is the
Candidate Elimination/Version Spaces learning.
>
>Also, you may want to try using Genetic Algorithms that take into
consideration combinations of attributes in order to select the best ones or
their combinations.
>
>I am not sure if these are available in Weka but these are the one you may
want to try first. I would also like to know if these are in Weka.
>

Quick answer:

Yes they are!
See e,g, Supervised Attribute Filters
where you have to choose an evaluator and
a search algorithm (which can be the Genetic algo),
or the attribute selected classifier wher the same choices are to be made
but resukts in a classifier with the selected attributes.


Hans van Rijnberk



>Regards,
>
>Subrat
>
>
>
>
>
>-----Original Message-----
>From: [hidden email]
>[mailto:[hidden email]]On Behalf Of rich
>Sent: Thursday, July 21, 2005 6:01 PM
>To: [hidden email]
>Subject: [Wekalist] classifier for working out which combinations of
>attributes give lowest error rate
>
>
>Hi,
>
>I have a dataset that i've been running through weka. Firstly I used 1R
>to determine the least error attribute. Then I ranked the arritbutes
>using the default ranker search method. I then used
>the attribute evaluator InfoGainAttributeEval to rank each attribute in
>terms of the info gain that each give.
>
>
>But, can I do this on combinations of attributes? Is there anything that
>I can run that looks at combinations to see which attributes clump
>together well to add to the information gain? Maybe something like
>1R,but  that instead of working on a single attribute can work on more
>that one so i could find which 2 attributes when combined produce the
>lowest error rate?
>
>thanks
>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
>
>
>


Hans van Rijnberk

[hidden email]



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