I am using the weka AttributeSelection filter on a dataset. Currently I
am using ClassifierSubsetEval or WrapperSubsetEval in combination with
different Classifiers and different search methods (often geneticSearch).
I have two aims:
On the one hand I want to output an arff-file with selected attributes.
But on the other hand I also need a report of how good this new
attribute subset performs (For example the value the geneticSearch uses
to decide which subsets to keep and which ones to discard). I need this
to compare the outcomes of different attribute selections.
Best would be to output the evaluation for every time a new subset is
tested during the (Genetic)Search. But it would be also ok if I would at
least have the last performance value when the selection is finished.
Can I accomplish this using the command-line interface?
Or do I have to use the API?
If so where should I start since I am not that deep into Java?
Dear Weka developer,
I have this problem with WrapperSubsetEval class:
I am using WrapperSubsetEval with LibSVM classifier and GreedyStepwise but I
am getting different results when I use the same method in GUI than when I
use it inside my Java code. This is my java code:
int indices; int i;
AttributeSelection Filter = new AttributeSelection ();
WrapperSubsetEval eval = new WrapperSubsetEval();
GreedyStepwise search = new GreedyStepwise();
Classifier svm = new LibSVM();
String newfeatures = Utils.arrayToString(indices);
I got the following results with diabetes dataset: 2,7,9
while when I use weka explorer I get different results.
I have included the picture of the Weka explorer with the
results.<nabble_img src="wrapper.jpg" border="0"/>
So, can you tell me where the error is please?
View this message in context: http://weka.8497.n7.nabble.com/WrapperSubsetEval-GUI-and-Java-code-results-…
Sent from the WEKA mailing list archive at Nabble.com.
I've implemented a new classifier, J48Consolidated, based on J48
(J48Consolidated extends J48) and it seems it works fine when is used from
Explorer or Experimenter.
I've tried to build a package for the package management system and, even
thought it seems it is installed OK, I can't choose our algorithm to use
it. In fact, the 'trees' fold appears empty.
I checked the weka.log file and I found the next error:
Exception in thread "AWT-EventQueue-1" java.lang.IllegalAccessError: tried
to access field weka.classifiers.trees.J48.m_Seed from class
I had to change the visibility of some members of the J48 class (from
private to protected) to be able to access them from J48Consolidated class.
One of this was m_Seed atribute.
In order to build the package, I also compiled J48 class and included it
within the .jar file and I got the mentioned error.
Can I build a package with a new classifier if I changed an original class
How should I build the .jar file for the package in this case?
Can someone help me with this issue, please?
Thank you in advance
I am performing a task of text classification using weka. I have created a
model(Random Forest) from weka using the training dataset given to me. I am
using this model in java to classify the text on test data which is going to
For the purpose of testing my code I have generated 500 positive and
negative examples. There are two ways I tried predicting on this test set.
One by loading the entire 500 examples, creating a single .arff file and
predicting against the model(from training). This gives good predictions for
the test set.
Another is to load individual examples and predict against the model. This
is not giving the same results as with the previous one. But this is how I
would want to use.
I am not able to figure out the reason this behavior. What is the difference
between the two? Is there any way to accomplish this.
View this message in context: http://weka.8497.n7.nabble.com/Different-predictions-on-the-same-test-datas…
Sent from the WEKA mailing list archive at Nabble.com.
I want to do 10 fold cross validation manually because i want to apply SMOTE only on training set in every iteration of cross validation, and then produce one evaluation of the 10 fold cross validation.
I've seen the code of the weka.classifiers.evaluation.Evaluation.crossValidateModel but i can't understand where or how it combines the 10 models that are created and produces one total evaluation.
Any help about how i could that?
Should i add some lines in weka.classifiers.evaluation.Evaluation.crossValidateModel source code to do SMOTE when i call this routine?
Thank you in advance.Divolis Alexandros
I am a student at the Autonomous University of Chihuahua, Mexico and I am new to the use of WEKA. My question is, why the equations obtained in the regression models generated in Weka are different from those obtained in excel charts? For example:
Thank you in advance.
JC Betancourt Ruiz
Thanks! I downloaded the latest package release, and I can confirm that it
is now working with my sparse dataset.
Date: Tue, 01 Apr 2014 16:57:02 +1300
From: Eibe Frank <eibe(a)waikato.ac.nz>
Subject: Re: [Wekalist] Problems with OPTICS using sparse data
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
It turns out that handling of sparse data wasn't implemented correctly
in OPTICS and DBScan. We have just made a new package release (1.0.4)
for WEKA 3.7 where this problem is fixed. The fix will also be
incorporated into WEKA 3.6.
Thanks for the bug report!
This is a repost as i couldnt get a reply [My initial post was on 25
Kindly help me with these 2 queries
1- can any one guide me with the name of any Offline tool which can convert
json files to arff format . There are some online converter available but i
can't use them as my file is very large in size so i need an offline
2- Please guide me ,where i can find various theoretical algorithms which
is used in Weka