many classifiers output a confidence score or probability score when
outputting predictions. With classifiers such as SVM's the confidence
scores or predictions add up to 1 I believe.
Is there a classifier where the confidence scores don't total 1? The
reason being is that I have 2 classes for my dataset yet there is a
possibility that an instance may belong to neither of the classes. I
cannot simply add a third class as I have almost no instances for training.
I want to be able to say that if a confidence score is low for each of
the classes I can assume that the instance belongs to neither of the
classes in the dataset. I can't do this with SVM's as the confidence
scores for each class total one. ie instance belongs to class1=0.8 and
instance belongs to class2=0.2
>Date: Wed, 13 Sep 2006 10:17:50 -0700
>From: "Xinghuo Zeng"
>Subject: Re: [Wekalist] How to use other distance/similarity measure
> in XMeans
>Thanks so much for your reply.
>I tried the command line and use "CosineDistance" I implemented. It
>does not work. The error message is that "weka.core.EuclideanDistance
>is not assignable from weka.core.CosineDistance". Do I need to modify
>the code of XMeans to use the cosine distance?
Peter has just recently posted a fix in the CVS. I think this wouldn't
be a problem in the latest version message.
Hi, I was wondering if there was any development going on for any sort
of probabilistic decision trees, specifically ones that use probability
bagging as described by Provost & Domingos "Tree Induction for
Probability-Based Ranking" from Machine Learning Vol. 52, #3, 9/2003.
- Michael C. Piantedosi
I was wondering the threshold value for the correlation coeficient of
two attributes to be filetered out. I have two attributes with
correlation coeficient of 0.62 but the correlated feature subset
selection method is selecting both of them. If I understood the
documentation correctly, this method selects the attributes that are
highly correlated to the label but least correlated to other attributes
in the subset.
Zentrum für Informationsdienste und Hochleistungsrechnen (ZIH)
Technische Universität Dresden Tel: (+49) 351 463-38776
Room 1019 Fax: (+49) 351 463-38245
Noethnitzer Straße 46
Is thery any automated procedure or tool to convert the data in arff/csv format to format that supported by libsvm ie <label> <attribute>:<value>...
kindly help me.
Thanks in Advance
Get your email and more, right on the new Yahoo.com
I was wondering if anyone could tell me how to use other distance in XMeans
(clustering algortihm)? It seems that the XMeans class only accept the
distance class "EuclideanDistance".
Can anybody explain how to perform one-class svm clasification in WEKA.
I have weka 3-5-3 version.It's working good.
My work is on FRAUD DETECTION and my data set contains 33 attributes out of which one is fraud lable(fraud found or not found).
How to apply 1-class svm to that data set?
Do I need to remove the fraud label from the data set/not?
I have divided the data set into two parts.
Part1 contains entire fraud data(ie fraud label is yes) and part2 contains entire non fraud data(ie fraud label is no).Two parts has 33 attributes and order of them is same.
Can anybody suggest me the solution.
My main problem is I dont know how to set the training and test data into WEKA for 1-class svm classifications.
Thanks in Advance,
All-new Yahoo! Mail - Fire up a more powerful email and get things done faster.
I have been using weka.associations.Apriori from a java program. I have
managed to use Apriori.buildAssociations to generate rules and to print
them using the "toString" method.
I would like to do some postprocessing on the rules. However, I could
not find any methods for extracting individual rules to check their
antecedents and consequents. For example, I would like to do the
following kinds of operation:
- get all rules that mention a particular attribute or a particular
value for an attribute and optionally state whether it should be an
antecedant or consequent.
- get a rule at a certain position (first, second etc.) and
get its antecedants or consequents;
Is there a way to do this?
I am a postgraduate now studying Data Mining by Weka.
today,when I run the source code "Java program for classifying short
text messages into two classes" ,Also the sample code appeared in your book
as well, Now I meet problems by -t option, I don't know which kinds of file
i need input ? Is the file type is .class file
Can you just give me a example how to run the program ? I mean give two
sample files, and the command line , as well as the results.
Waiting for your reply!
Thanks a lot !
University of WuHan China
In order to modify the Genetic Search algorithme in weka system, I want to modify the class attributeSelection.GeneticSearch. but I don't know how to compile all the weka classes to generate a nother version of weka systm contining my modification.