what do you think about using
first a variable reduction method ?
In weka - select attributes you have the
possibilty (..i like the ranker method)
identify important/unimportant variables for the classification variable.
....and after this using part.Part ( incl. pruning) with
a subset of your variabes which extract you important rules.
( binary classification variable 1/0 you have recode to bad/good or something else)
"Playing" with different settings of SubsetEvaluators
and the part.PART parameters and observe the different results
is not unimportant !?
Sorry for the confusion. I was asking about possibility to evaluate rules
(in prepositional clause format) by entering them directly to WEKA. Simply
speaking does WEKA contain a rule interpreter that can be used as a
If it doesn't can you point me to some other tool that has this ability and
is simple to use?
> -----Original Message-----
> From: C. Setzkorn [mailto:C.Setzkorn@csc.liv.ac.uk]
> Sent: Monday, September 23, 2002 5:41 PM
> To: Szymon Kobalczyk
> Cc: 'wekalist(a)list.scms.waikato.ac.nz'
> Subject: Re: [Wekalist] Evaluating symbolic rules
> This might be relevant:
> author = "Nada Lavrac and Peter Flach and Blaz Zupan",
> year = "1999",
> title = "Rule Evaluation Measure: A Unified View",
> series = "Lecture Notes in Computer Science, Lecture Notes in Artificial
> Inteligence, 1634",
> pages = "pp. 174-185",
> publisher = "Springer",
> url = "http://www.csc.liv.ac.uk/~chris/registred_papers/4.zip"
> You can also determine the AUC (area under the ROC curve) of a rule
> Szymon Kobalczyk wrote:
> > Hi,
> > Is there a way to evaluate a set of symbolic rules, that I enter
> > on a given data set?
> > I've generated some rules and then fine tuned them by hand. Now I want
> > check how good they are.
> > Szymon Kobalczyk
> > _______________________________________________
> > Wekalist mailing list
> > Wekalist(a)list.scms.waikato.ac.nz
> > http://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> All the best
> Mr. C. Setzkorn (PhD Student)
> Department of Computer Science
> University of Liverpool
> Chadwick Building, room G45a
> Peach Street, Liverpool L69 7ZF
> United Kingdom
> Email: chris(a)csc.liv.ac.uk
> Phone: 0044 151 794 3694
> Fax: 0044 151 794 3715
> homepage: http://www.csc.liv.ac.uk/~chris
ok, i try next few weeks some attempts
how it is easy to use one data-file for both
weka and R !
My first idea is using MySql with JDBC, because
R have an RODBC and RMySQL package, and RODBC works well !
..causing a job change, i didn't
know, how long i have time for this,
but as i'm mentioned (in another posting)
a PH.D or normal university project offer
in Germany/Berlin would be very nice ;-)
I wouldn't say that I know R -- I am just a user. However, a direct link
Weka <-> R is certainly an excellent idea. At the moment, for each and
every experiment, I run Weka from the command line and redirect the
output to a cvs file, which I then load into R for statistical analysis,
summarizing, and (especially)
plotting. The thing is, I wouldn't really know how such a link would be
supposed to look like -- the thing I like most about R is its
flexibility, that you script whatever you want with your data.
I have been using R as an end-user for three years, and as a developer for
almost 2 years. I have been thinking of implementing a RWeka package, which
will allow R users to invoke Weka within R and export/import data back and
forth. (kind of like Brian Ripley's xgobi package for R).
However I haven't got time, at the moment, to even start on this...
I have a few questions here.
1. I want to build a tree to make numeric predictions. All attributes in
my training dataset are numeric (infact binary). There are quite a
few Classes in the Classifiers package. Which is the best Class I
can use for this purpose?
2. My data set is huge, (5807x693) and I am using the M5Prime Class to
build a few trees, each on a different set of attributes of this dataset.
I found that with each tree built, 40MB more memory is being used up. Does
a tree actually take up so much memory. If not, how can I reduce the
amount of memory being consumed?
Thanks in advance,
Is there a way to evaluate a set of symbolic rules, that I enter manually,
on a given data set?
I've generated some rules and then fine tuned them by hand. Now I want to
check how good they are.
Experimenting with WEKA I came to the need of automatically generating
trainingsets of data for
generating a model and also producing autoamtically a hold-out testset for
validating the created models.
Thus creating a automated walk-forward train- and out-of-sample test series
How can this be done, without the need to manually create these train- and
Can this be done in the Experimenter? If yes, then how?
Perhaps it can be done with CLI-commands. I couldn't figure it out myself,
so I hope somebody can help me!
Please send a copy of your response to my emailadress : jfhm_bours(a)planet.nl