[prev in list] [next in list] [prev in thread] [next in thread] 

List:       wekalist
Subject:    [Wekalist] Re: Greetings from Iceland
From:       Mark Hall <mhall () cs ! waikato ! ac ! nz>
Date:       2005-11-28 18:58:26
Message-ID: a96c12e50511281058n733eeffbt948f7efa3a0bc19e () mail ! gmail ! com
[Download RAW message or body]

Hi Tony,

PrincipalComponents is indeed an unsupervised method. Unfortunately it
doesn't really fit all that well in the Explorer with Weka's other
attribute selection methods (which are supervised). If a class
attribute has been specified for the data then PC will ignore that
attribute in computing the components. In the Explorer this is set by
default to the last column in the data, unless you change it using the
drop down box. In many cases this is actually useful. For example,
when you combine PC with a classifier and want the class attribute
ignored for the transformation but passed on to the classifier.

You can achieve full unsupervised operation with PC if you run it from
the command line. If no class attribute is specified on the command
line then PC will use all the attributes when computing the
transformation.

Best regards,
Mark Hall.

On 11/29/05, Tony Chan <tony@unak.is> wrote:
>
>
>
> I am a teacher at the University of Akureyri in Iceland. Merry Christmas
> from the north!
>
>
>
> We are using WEKA here for our students. Thank you for your work on
> PrincipalComponents. I have one comment. PCA belongs to the class of
> attribute selection method call unsupervised. This means that the user does
> not provide class label information, e.g., the Alate Adelges data from
> http://mo161.soci.ous.ac.jp/@d/DoDStat/alate/alate_dataE.xml.
> This dataset consists of 40 alate adelges (winged insects) and 19 variables
> which measure body parts.
>
>
>
> I click the "Select attributes" button on this data set, then "Choose"
> PrincipalComponents Evaluator and Ranker method. Now on the left, near the
> middle, attribute V19 has been automatically chosen as the class attribute.
> In this case, there is no class attribute being supplied or needed for PCA.
> That is, PCA should be performed on all 19 attributes.
>
>
>
> Let me know what you think. Thanks.
>
>
>
> Tony
>
>


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

_______________________________________________
Wekalist mailing list
Wekalist@list.scms.waikato.ac.nz
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist

[prev in list] [next in list] [prev in thread] [next in thread]