Allen Harper | 27 Jan 21:29 2015
Picon

AttrributeSelectedClassifier with Training/Testing pairs

Hi Everyone,

I am working on a program that will process 10-element parallel Instance
arrays containing matched training and testing sets. Attempting to perform
various attribute selection methods on each pair per learning algorithm. 

I've adapted the CV code that I found on the Use WEKA in your Java code
section. My code looks like the following.

The issue I am having is that the evaluateModel method appears to be
throwing a null pointer exception.

Can someone help me fix this issue...

Thanks, Allen

	public static Evaluation classify(Classifier model, Instances trainingSet,
			Instances testingSet) throws Exception {

		String[] optionsW = new String[4];
		optionsW[0] = "-F";
		optionsW[1] = "5";
		optionsW[2] = "-T";
		optionsW[3] = "0.01";

		String[] optionsS = new String[4];
		optionsS[0] = "-D";
		optionsS[1] = "1";
		optionsS[2] = "-N";
		optionsS[3] = "5";
(Continue reading)

double d s | 27 Jan 19:40 2015
Picon

enhance classification accuracy

Hi,

My accuracy result was very bad, so I performed several techniques to improve it, such as invoking different classification algorithms or tuning their parameters or applying Weka  methods to remove outliers or performing attribute selection, but after all, the Correctly Classified Instances  =  29.3796 %.

Is there any further suggestions to enhance classification accuracy?

Thanks.
Sandler
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Ashley Painter | 27 Jan 18:24 2015

Splitting Training testing data

Hi,
I was wondering how when you select for WEKA to split data by a percentage into training and testing data when it runs an analysis that it actual does so. Is it random or does it just take the first say if you had selected 70% of the data points then the other part?
Sincerely,
Ashley
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Yaakov HaCohen-Kerner | 27 Jan 18:06 2015
Picon
Picon

Which unsupervised ML methods... to choose when dealing with clustering?

Dear all,

Could you direct me to papers or documents that advise concerning
which unsupervised ML methods, parameters and their values, features and how many requested cluster categories to choose when dealing with various clustering tasks?

Best!
Yaakov
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Yaakov HaCohen-Kerner | 27 Jan 18:03 2015
Picon
Picon

How can we implement the produced models in our programs?

Dear all,

(1)
Most of the ML methods (e.g. SVM and MLP) produce models that are "black boxes".
How can we implement these produced models in our programs?

(2) Is there any variant of LIBSVM in WEKA?


Best!
Yaakov
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Yaakov HaCohen-Kerner | 27 Jan 17:57 2015
Picon
Picon

10-cross validation on the training set or on the test set?

Dear all,


I know what is 10-cross validation.

What is the meaning of 10-cross validation on the training set
versus 10-cross validation on the test set?

Should we do both on only one of them (which one is more important)?

Best!
Yaakov
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Yaakov HaCohen-Kerner | 27 Jan 16:37 2015
Picon
Picon

How to run complex and multiple jobs via the command-line?

Dear all,


I understand that I can save time if I run via the command-line instead of run via the regular menu.
Where can I read about how to use the command-line and how to run complex jobs, e.g.: run of a few ML methods (with default values of the parameters and even with certain values) on the same input ARFF/CSV table (or even on a few tables)?


Best!
Yaakov
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Yaakov HaCohen-Kerner | 27 Jan 16:28 2015
Picon
Picon

Which ML method(s) to choose when we want to classify/cluster (...)?

Dear all,

Can you direct me to papers (or other documents) that advise which ML method(s) (and maybe with which parameters' values) 
to choose when we want to classify/cluster 
as a function of the number of 
  • # of features
  • # of classes (categories)
  • # of rows (representing documents)
  • type of classification/clustering task
  • types of features (e.g., POS-TAGs, averages, ...)

Best!
Yaakov
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Yaakov HaCohen-Kerner | 27 Jan 16:20 2015
Picon
Picon

What to do if a certain input document belongs to several classes?

Dear all,


Let us assume that a certain input document (each document is represented in one line in the ARFF/CSV table)
belongs to several classes.

Should I duplicate the line (row) representing this document
in the ARFF/CSV table according to # of classes that the document belongs to them
while changing the value of the correct class at the last column for each line? 


Best!
Yaakov
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Yaakov HaCohen-Kerner | 27 Jan 16:09 2015
Picon
Picon

How can I know that my best results are significantly better than other results?

Dear all,


Reviewers are asking from time to time, whether my best results are significantly better than other results and I don't know which measures (or other thins) can help me to answer their questions.

And now formally,
If for a certain classification task I got the best results using a specific ML method (let us call it X) versus other ML methods, how can I know whether the results of X are significantly better than those of the other ML methods?


Best!
Yaakov
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Yaakov HaCohen-Kerner | 27 Jan 15:47 2015
Picon
Picon

How can I do automatic tuning of the parameters of any ML method?

Dear all,

If it is possible - how can I do automatic tuning of the parameters of any ML method?

If it is not possible - how do you recommend me to to do the tuning process in a systematic way?

Best!
Yaakov

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html

Gmane