gmane.comp.ai.weka

Re: Generate association rules

scott mclelland — 2013-05-24T09:32:11

Take me off this!! I unsubscribed but still get all the emails !

Sent from my iPhone

On 24 May 2013, at 04:18, "Mark Hall" <mhall< at >pentaho.com> wrote:

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

Re: INTREPETING RESULTS OF GENETIC SEARCH

Mark Hall — 2013-05-24T03:29:42

From: priyanka chelladurai <priyadio< at >gmail.com<mailto:priyadio< at >gmail.com>>
Reply-To: "Weka machine learning workbench list." <wekalist< at >list.scms.waikato.ac.nz<mailto:wekalist< at >list.scms.waikato.ac.nz>>
Date: Thursday, 16 May 2013 5:49 AM
To: "Weka machine learning workbench list." <wekalist< at >list.scms.waikato.ac.nz<mailto:wekalist< at >list.scms.waikato.ac.nz>>
Subject: [Wekalist] INTREPETING RESULTS OF GENETIC SEARCH

Hi

I am basically doing text classification with SMO using weka. I an applying genetic search algorithm for attribute selection with wrapper subset evaluator.. I get the following output for genetic search after which the selected attributes are displayed.
I am not able to understand this output...

what does merit,scaled represent....?

Actually wrapper subset is evaluating each subset of attributes using a classifier with 5 fold cross validation which is basically the fitness function for genetic search. so which parameter in the output represent that fitness score?

I have set number of generatio

Re: A question about weka and its java source codes

Jessica Beltrán — 2013-05-24T03:28:17

I did something like that but in my case I evaluated instance by instance,
something like this:

 for(int ins=0;ins<numInstancias;ins++){
   vector = testData.instance(ins);
   claseIn = vector.value(nFeatures);
   claseOut = eval.evaluateModelOnce(libsvm, vector);
}
evaluateModelOnce returns the prediction made by the clasifier.

I hope this work for you.

-Jessica



2013/5/23 李宁 <leening0807< at >gmail.com>

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

Re: Generate association rules

Mark Hall — 2013-05-24T03:16:44

From: "amanda< at >ctim.mar.mil.br<mailto:amanda< at >ctim.mar.mil.br>" <amanda< at >ctim.mar.mil.br<mailto:amanda< at >ctim.mar.mil.br>>
Reply-To: "Weka machine learning workbench list." <wekalist< at >list.scms.waikato.ac.nz<mailto:wekalist< at >list.scms.waikato.ac.nz>>
Date: Wednesday, 15 May 2013 2:35 AM
To: "Weka machine learning workbench list." <wekalist< at >list.scms.waikato.ac.nz<mailto:wekalist< at >list.scms.waikato.ac.nz>>
Subject: [Wekalist] Generate association rules

Hi,
I have a large dataset with only numbers and I need take rules of association with the number of defects. I need do An Investigation of the Effect of Module Size on Defect Prediction.
In my search I read that Apriori don't work with numeric atributes only nominal. My dataset are only with numeric atributes.
My big problem is that I need something that I don't miss the precision in my numbers. I had discretized the Numeric Attributes into Nominal but it didn't served me because occured this problem.
Can somebody help me with the best way to do this association with

Re: Output predictions in csv format

Eibe Frank — 2013-05-24T02:29:23

You need WEKA 3.7, this isn't available in 3.6. Just add 

  -classifications "weka.classifiers.evaluation.output.prediction.CSV"

somewhere before the double hyphen. 

Cheers,
Eibe

On 24 May 2013, at 09:48, Arik Harel <harel< at >marine.rutgers.edu> wrote:



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

Re: Ignore (or skip) an attribute fromweka.classifiers.meta.AttributeSelectedClassifier

Eibe Frank — 2013-05-24T02:24:44

You could wrap an AttributeSelectedClassifier into a FilteredClassifier, where the Remove filter is used as the filter in the latter classifier.

Cheers,
Eibe

On 24 May 2013, at 10:59, Arik Harel <harel< at >marine.rutgers.edu> wrote:



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

A question about weka and its java source codes

李宁 — 2013-05-24T00:44:06

Hello, all
      I am using weka.jar and libsvm.jar to develop my project. When I want
to output the predictions, I do not find the functions to satisfy my
requirements. I know I can easily get the prediction list with weka
software, just with the Classify ---> More options --> Output predictions
-->PlainText. But I want to know how I can get the list with weka,jar.
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist< at >list.scms.waikato.ac.nz
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html

Ignore (or skip) an attribute fromweka.classifiers.meta.AttributeSelectedClassifier

Arik Harel — 2013-05-23T22:59:43

Hi,
I would like to use ID attribute in the result predictions table, and I
would like the classifier to ignore this ID attribute.

I have looked at: http://weka.wikispaces.com/Instance+ID
However the suggested option "
-Fweka.filters.unsupervised.attribute.RemoveType" does not work for
weka.classifiers.meta.AttributeSelectedClassifier, or at least I have no
idea how to make it work.

The first column of the input files (train, test) contains ID.

This is my code (sh file):
java weka.classifiers.meta.AttributeSelectedClassifier \
        -t data/books_train_ids.arff \
        -T data/books_test_ids.arff \
        -c 2 \
        -p 1 \
        -E "weka.attributeSelection.ChiSquaredAttributeEval " \
        -S "weka.attributeSelection.Ranker -T 0.001 -N -1" \
        -W weka.classifiers.functions.SimpleLogistic \
        -- \
        -I 0 -M 500 -H 50 -W 0.0  \

* I have paseted in the bottom of the page the input files.

Please help.
Thank you,

Arik



This is the train file:
< at >relation
imaginaryBooks-weka.fi

Output predictions in csv format

Arik Harel — 2013-05-23T21:48:34

Hi,
How do I output my predictions in CSV format, using command line.

Here is my command line (it is inside an sh file):

java weka.classifiers.meta.AttributeSelectedClassifier \
        -t data/books_train.arff \
        -T data/books_test.arff \
        -c 3 \
        -p 0
        -E "weka.attributeSelection.ChiSquaredAttributeEval " \
        -S "weka.attributeSelection.Ranker -T 0.001 -N -1" \
        -W weka.classifiers.functions.SimpleLogistic \
        -- \
        -I 0 -M 500 -H 50 -W 0.0  \

I read somewhere that I can just add:
"weka.classifiers.evaluation.output.prediction.CSV"
or:
- classifications  "weka.classifiers.evaluation.output.prediction.CSV "

But I can't get it to work no matter where I stick it in the command lin
above.

Help.
Thank you!



Arik
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist< at >list.scms.waikato.ac.nz
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.wa

Re: Weka with Data Mining for Research

Eibe Frank — 2013-05-23T21:06:47

You can't load this kind of data into the Explorer (unless you have lots of RAM…), but you may be able to apply one of WEKA's incremental learning algorithms from the command-line or through the KnowledgeFlow interface. Given an appropriate evaluation method (i.e. no cross-validation), they don't load all the training data into memory; instead, they load only one instance at a time. Search for classifiers that implement UpdateableClassifier or UpdateableClusterer. Two methods that you should be able to run on your data are NaiveBayesUpdateable and SGD (stochastic gradient descent for learning linear models).

Alternatively, try MOA, which is a workbench for data streams. It has a larger variety of incremental algorithms. You can also use it through WEKA by importing the MOA package.

Cheers,
Eibe

On 24 May 2013, at 05:31, dirichlet <dirichlet89< at >gmail.com> wrote:



_______________________________________________
Wekalist mailing list
Send posts to: Wekalist< at >list.scms.waikato.ac.nz
List info and subscripti

Re: Re: Using data from an arraylist in Java

Eibe Frank — 2013-05-23T20:58:11

You can use the constructor in the DenseInstance class to make Instance objects from your data and use these for classification.

Cheers,
Eibe

On 24 May 2013, at 06:03, simmisj <simmisj< at >gmail.com> wrote:



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

Re: Using data from an arraylist in Java

simmisj — 2013-05-23T18:03:54

Nobody has an idea about this? :/



--
View this message in context: http://weka.8497.n7.nabble.com/Using-data-from-an-arraylist-in-Java-tp28069p28087.html
Sent from the WEKA mailing list archive at Nabble.com.

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

Re: Weka with Data Mining for Research

dirichlet — 2013-05-23T17:31:17

I'm also quite a newbie in data mining but I do feel that 750Gb is really what falls under the 'Big Data' label and can be considered to be something more `advanced' (thus maybe not appropriate for approaching data mining as starters)

In order to use Weka you could try some dimensionality reduction techniques as sampling and feature reduction after having, of course, analyzed your data  (you could find that those 750gb are referred to a time period of many years for example so you can choose to start concentrating only on one particular year), but maybe the information loss you will face is not adequate to your `business' goals.

Maybe someone here can elaborate more on the requirements of memory and cpu power in order to have a feasible data mining step.

Cheers,
farinadiceci

Il giorno 23/mag/2013, alle ore 19:20, Samith Dilshan Fernando ha scritto:


_______________________________________________
Wekalist mailing list
Send posts to: Wekalist< at >list.scms.waikato.ac.nz
List info and subscription status: htt

Weka with Data Mining for Research

Samith Dilshan Fernando — 2013-05-23T17:20:52

i'm a final year Bsc in IT student at SLIIT(Sri Lanka Institute of
Information Technology-http://www.sliit.lk/) in Sri Lanka.

For our final year we are doing a research project and my group has chosen
to do a project related with data mining. Data mining is a new concept for
sri lanka and even we couldn't find any person who done a project for data
mining.

My team selected to do our research by using "WEKA" for build our mining
model because weka is related with JAVA. personally i'm very interesting in
this research now and currently we are searching how to works with weka.
and we have to build data mining algorith as well. So  We are still
confusing in some areas how to do this project.

our research is "Failure prediction for decision makers in data centers
using data mining". So we are try to predict failures in data centers in
sri lanka becouse my country still not families with this huge concept. We
get a huge data set (Size is 750GB) from a leading organization in the IT
Industry.

problem is we foun

Re: A question about one class svm

Jessica Beltrán — 2013-05-23T16:49:47

As I know the database, I know that instance 4 is not 'Iris-setosa' I know
it comes from 'Iris-virginica' class, I just labeled as ? to use the
one-class classifier.


2013/5/23 李宁 <leening0807< at >gmail.com>

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

Re: Use of ROC curves

Mark Hall — 2013-05-23T10:31:26

From: Martin O'Shea <appy74< at >dsl.pipex.com<mailto:appy74< at >dsl.pipex.com>>
Reply-To: "Weka machine learning workbench list." <wekalist< at >list.scms.waikato.ac.nz<mailto:wekalist< at >list.scms.waikato.ac.nz>>
Date: Wednesday, 15 May 2013 10:06 PM
To: "Weka machine learning workbench list." <wekalist< at >list.scms.waikato.ac.nz<mailto:wekalist< at >list.scms.waikato.ac.nz>>
Subject: [Wekalist] Use of ROC curves

Hello

I’ve recently been using Weka to classify permutations of keywords from RSS feeds involving use of stemming, ngrams and so on using decision trees. So in order to determine the best result, I would like to use ROC curves based upon the numbers of TP / FP per permutation.

However, I have seven classes.

Can Weka generate ROC curves for this many classes on one plot? Or would I have to use an individual plot per class and average them at the end?

You'll probably have to do your own averaging of the curves.

Cheers,
Mark.

_______________________________________________
Wekalist mailing list
Send posts to: Wekali

Re: Query on Support in Apriori Algorithm

Excel — 2013-05-23T09:03:24

Hello,

   Any help on the topic?

Regards,
ER



--
View this message in context: http://weka.8497.n7.nabble.com/Query-on-Support-in-Apriori-Algorithm-tp28070p28081.html
Sent from the WEKA mailing list archive at Nabble.com.

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

Re: A question about one class svm

李宁 — 2013-05-23T08:21:03

In your letter, you have written "But you have to kown……". What is your
meaning?


2013/5/23 Jessica Beltrán <jessicabeltran< at >gmail.com>

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

Re: A question about one class svm

Jessica Beltrán — 2013-05-23T04:28:01

http://en.wikipedia.org/wiki/Receiver_operating_characteristic " It is
created by plotting the fraction of true
positives<http://en.wikipedia.org/wiki/True_positive> out
of the positives (TPR = true positive rate) vs. the fraction of false
positives <http://en.wikipedia.org/wiki/False_positive> out of the
negatives (FPR = false positive rate), at various threshold settings."

It's done for two classes, you need TP and FP.
With one-class classification in weka you don't get the False Positives
value in the explorer, because you don't know anything about the missing
labels values '?'

An example of what you get is this:

=== Predictions on test data ===

inst#,    actual, predicted, error, probability distribution
     1          ?          ?      +  *0
     2          ?          ?      +  *0    //True negative
     3          ?          ?      +  *0
     4          ?   1:Iris-set  +  *1    //False positive
     5          ?          ?      +  *0
     6 1:Iris-set 1:Iris-set         *1      //True Positive

Re: How to calculate Precision/Recall Breakeven Point(PRBE) and Mean Average Precision (MAP)

Mark Hall — 2013-05-23T04:27:33

From: ismail bilgen <ismbil< at >gmail.com<mailto:ismbil< at >gmail.com>>
Reply-To: "Weka machine learning workbench list." <wekalist< at >list.scms.waikato.ac.nz<mailto:wekalist< at >list.scms.waikato.ac.nz>>
Date: Tuesday, 14 May 2013 9:14 PM
To: "Weka machine learning workbench list." <wekalist< at >list.scms.waikato.ac.nz<mailto:wekalist< at >list.scms.waikato.ac.nz>>
Subject: [Wekalist] How to calculate Precision/Recall Breakeven Point (PRBE) and Mean Average Precision (MAP)

Hi,

I am using Weka with Java and I need to calculate Precision/Recall Breakeven Point (PRBE) and Mean Average Precision (MAP).

The Evaluation class give the areaUnderROC already as follows:

..
Evaluation eval = new Evaluation(randData);
..
System.out.println("areaUnderROC: " + eval.areaUnderROC(1));

Is there any practical way to calculate "Precision/Recall Breakeven Point" and "Mean Average Precision: MAP"

The development version of Weka has a plugin mechanism by which new evaluation metrics can be added to the system. See the documentation at:

http://we

How can I improve the accuracy of my model? (SMO, Cost Matrix, GridSearch, CostSensitiveClassifier, Accuracy)

Saiph Kappa — 2013-05-23T02:32:35

Hi,

I am trying to improve the accuracy of my model with SMO/SVM and RBFKernel.

I have a training set with 164 examples where 7 examples are true and the
rest are false.

Since I have a great imbalance, I defined a cost matrix like this:
«% Rows  Columns
2       2
% Matrix elements
0.0     4
96      0.0»

The problem is that I was unable to use the meta classifier GridSearch (to
find the parameters C and Gamma) with the CostSensitiveClassifier. Would it
be possible to use those 2 meta classifiers together? Or the parameters C
and Gamma do not depend on the cost matrix?

After testing my model with a test-set of the same size, but with different
examples, I got the following result:
http://pastebin.com/vga89yH2(training and test set are also included).

Basically, it starts giving me true before the time is right. Example:
«
Actual : Predicted

[0]:[0.0]
[0]:[0.0]
[0]:[0.0]
[0]:[1.0]
[0]:[1.0]
[1]:[1.0]
...
»

Would it be better if I transform the input, which is numerical, to
intervals (i.e. discrete v