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

List:       wekalist
Subject:    [Wekalist] Questions about BayesNets and Logistic Regression
From:       Jacob Scott <jhscott () MIT ! EDU>
Date:       2005-11-29 1:30:55
Message-ID: 20051128203055.2jita42txgggc8ok () webmail ! mit ! edu
[Download RAW message or body]

Hi,

I'm using some Weka for some run of the mill compbio tasks and had a few
questions about what I could do with and without hacking Weka source.

In particular I'm doing classification and regression with the following ARFF
file:

@RELATION ppi
@ATTRIBUTE intcount REAL
@ATTRIBUTE pearson REAL
@ATTRIBUTE clustering REAL
@ATTRIBUTE class {1,0}

@DATA
....

Here class is whether or not proteins interact... it's binary here but is really
a probability, based on the other attributes.

Questions:

1) Using NaiveBayes and Logistic classification, I am forced to have my class be
nominal. I'm looking for some way to say, set my own classification threshold
(true iff Pr(true) > .75). I'm pretty sure I can yank the model that Weka
outputs, run my data through it in matlab or python, and get the same place,
but if there's some flag somewhere (or some way to do logistic regression
rather than classification) that would be very helpful.

2) Using BayesNet regression on this data, if I don't first apply a Discretize
filter, I get an error. I've seen it mentioned before on the mailing list
without a solution, so I'm attaching the details below. I'm wondering if there
is an easy way to force a full bayes net instead of having Weka learn the
structure, or if I am stuck writing a BIF file?

Thanks very much,

Jacob

Trying to run BayesNet under classification tab on the above ARFF file I get the
following Error:

A nominal attribute (clustering) cannot have duplicative labels ('(0-0]').

My stack trace is:

Warning: discretizing data set
java.lang.IllegalArgumentException: A nominal attribute (clustering) cannot have
 duplicate labels ('(0-0]').
        at weka.core.Attribute.<init>(Unknown Source)
        at weka.core.Attribute.<init>(Unknown Source)
        at weka.filters.supervised.attribute.Discretize.setOutputFormat(Unknown
Source)
        at weka.filters.supervised.attribute.Discretize.batchFinished(Unknown So
urce)
        at weka.filters.Filter.useFilter(Unknown Source)
        at weka.classifiers.bayes.BayesNet.normalizeDataSet(Unknown Source)
        at weka.classifiers.bayes.BayesNet.buildClassifier(Unknown Source)
        at weka.gui.explorer.ClassifierPanel$17.run(Unknown Source)


_______________________________________________
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]