WEKA machine learning software discussion

Ashraf M. Kibriya | 1 Sep 2005 02:23

Re: plese help me...urgent

>
>
>> Hi, I'm a student of a Bicocca University in Milan, Italy and is the 
>> first time that I use Weka.
>> 
>> I want to create a text classifier for some documents of my University.
>> 
>> I have got only a database of 142 entry both for train and 
>> test.(multilabel problem with around 100 categories)
>> 
>> I have made the preprocessing phase myself, because my documents are 
>> written in italian and this language is very difficoult to preprocess 
>> with a automatic tool.
>> 
>> Now I have built all file in arff format, and I want to ask if there is 
>> some algorithm that is best for my problem. Now I have used Multiclass 
>> in order to aggregate my file and a neural network, but it take more 
>> than two hour to make a model...It's normal?
>  
>
Try using Multinomial Naive Bayes or the Complement Naive Bayes, they are generally more well suited for
text classification problems. For higher accuracy, you might want to try SVM (i.e. SMO classifier in
Weka), although it would take longer to run.

Regards,
Ashraf

headers

Ricardo Farfán | 1 Sep 2005 04:54

what data conform each cluster in K mean algorithm

Hello,

I have a java implementation of clustering analisis, I'm working whit K mean,

but I want to know what data conform each cluster in K mean algorithm

ATT: Ricardo Farfán

_______________________________________________
Wekalist mailing list
Wekalist <at> list.scms.waikato.ac.nz
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist

headers

Peter Reutemann | 1 Sep 2005 05:15

Re: distributing process

> I am trying the distributing process of the weka´s experimenter. But I 
> am having difficulties.Which the step that I must follow? 

You can find a step-by-step HOWTO for *nix on the Wiki:
   http://weka.sourceforge.net/wiki/index.php/Remote_Experiment

HTH

Cheers, Peter
--

-- 
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/     +64 (7) 838-4466 Ext. 5174

headers

Stefan Kuhn | 1 Sep 2005 19:43

programming problem

Hi all,
I'm using weka j48 in a webapp. I'm building the model from time to time. For 
this, I add all my training instances to an Intances object. When I later 
want to classify something, it seems I have to da a setDataset with my 
Instances object from the training on the instance to classify. For me, this 
means I have to hold a pointer to Instances in memory all the time. Since I 
have 352 attributes and some thousand instances, I want to avoid this. Is 
this possible? I would like to only hold the J48 object in memory.
Thanks,
Stefan
--

-- 
Stefan Kuhn M. A.
Cologne University BioInformatics Center (http://www.cubic.uni-koeln.de)
Zülpicher Str. 47, 50674 Cologne
Tel: +49(0)221-470-7428   Fax: +49 (0) 221-470-7786
My public PGP key is available at http://pgp.mit.edu

_______________________________________________
Wekalist mailing list
Wekalist <at> list.scms.waikato.ac.nz
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist

headers

David | 1 Sep 2005 19:30

Re: programming problem

Hello,

I simply delete the instances when I want to shrink the model:

ex.:

private void Shrink() {
m_TrainingSet.delete();
m_TestingSet.delete();
}

This remove all instances while the dataset "prototype" (features description) remains intact.

For logging purpose, you can backup instances by dumping them in an .arff file on disk.

public static void PrintToArffFile(WekaLearner pi_wl, String pi_filename, int pi_Set) {
  ArffSaver as = new ArffSaver();
  try {
   File fOut = new File(pi_filename);
   FileOutputStream os = new FileOutputStream(fOut);
   as.setDestination(os);
   as.setFile(fOut);
   if (pi_Set == PRINT_TRAINING_SET) {
    as.setInstances(pi_wl.GetTrainInstances());
   } else if (pi_Set == PRINT_TESTING_SET) {
    as.setInstances(pi_wl.GetTestInstances());
   } else {
    throw new Error("You must specify whether to output training or testing set.");
   }
   as.writeBatch();
  } catch (Exception e) {
   throw new Error("Unable to output the weka' arff file");
  }
}

David

On 9/1/05, Stefan Kuhn <stefan.kuhn <at> uni-koeln.de> wrote:

Hi all,
I'm using weka j48 in a webapp. I'm building the model from time to time. For
this, I add all my training instances to an Intances object. When I later
want to classify something, it seems I have to da a setDataset with my
Instances object from the training on the instance to classify. For me, this
means I have to hold a pointer to Instances in memory all the time. Since I
have 352 attributes and some thousand instances, I want to avoid this. Is
this possible? I would like to only hold the J48 object in memory.
Thanks,
Stefan
--
Stefan Kuhn M. A.
Cologne University BioInformatics Center ( http://www.cubic.uni-koeln.de)
Zülpicher Str. 47, 50674 Cologne
Tel: +49(0)221-470-7428 Fax: +49 (0) 221-470-7786
My public PGP key is available at http://pgp.mit.edu

_______________________________________________
Wekalist mailing list
Wekalist <at> list.scms.waikato.ac.nz
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist

--
Balie - Baseline Information Extraction
http://balie.sourceforge.net
[Open Source ~ 100% Java ~ Using Weka ~ Multilingual]

_______________________________________________
Wekalist mailing list
Wekalist <at> list.scms.waikato.ac.nz
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist

headers

Peter Reutemann | 1 Sep 2005 22:43

Re: Prediction

> I am having a problem calling Weka from the emacs shell on my Windows 
> machine. I have read the notes found at http://alex.seewald.at/WEKA/ . I 
> think it is a CLASSPATH setting problem.

I'm normally working with cygwin under Win32 or on a Linux box.

> Working from the directory to which I downloaded the file 
> callClassifier.class <http://alex.seewald.at/WEKA/callClassifier.class>. 
> I get:
> 
> *>set CLASSPATH .:$Weka-3-4/weka.jar*
> 
> set CLASSPATH .:$Weka-3-4/weka.jar
> 
> Environment variable CLASSPATH not defined
...

Since you're using windows, why don't set up the classpath there?
- right-click on "My Computer"
- select "Properties"
   (the alternative is via the Control Panel and then "System")
- under WinXP go to the "Advanced" tab
- click on "Environment Variables"
- create a new user variable (or system-wide for all users if you
   prefer): CLASSPATH
- and add the following value (assuming that Weka is installed in
   "c:\program files\weka-3-4")
   .;c:\program files\weka-3-4\weka.jar
- close all the dialogs with "OK"

Running Weka from command line:
- Start -> Run... and enter "cmd" followed by <Enter>
- e.g., run this command to see J48's help:
   java weka.classifiers.trees.J48 -h

HTH

Cheers, Peter
--

-- 
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/     +64 (7) 838-4466 Ext. 5174

headers

Eibe Frank | 1 Sep 2005 22:45

Re: Very different results in 2 weka SMO versions?

No, they shouldn't be viewed that way.

Cheers,
Eibe

On Aug 31, 2005, at 5:41 PM, yee seng chan wrote:

> Eibe, thanks for the prompt reply! I've 1 more question below:
>  
>> The reason is that the old version used the standard sigmoid function
>> (i.e. without any fitting) to transform the SVM's output into [0,1]. 
> So this sigmoid function is not parameterized (i.e. not fitted to the 
> training data). In this case, can the real numbers output from this 
> old SMO version be considered to be proper probability estimates?
>  
> Thanks,
> Yee Seng.
>
>> In
>> default mode, the new version computes a "probability" based on the
>> number of votes each class receives in pairwise classification. To get
>> proper probability estimates you need to turn on the logistic model
>> option, which fits a logistic model (i.e. parametrized sigmoid) to the
>> output of each SVM in the pairwise classifier and then performs
>> pairwise coupling.
>>
>> Cheers,
>> Eibe
>>
>> On Aug 31, 2005, at 3:32 AM, yee seng chan wrote:
>>
>> > Hi,
>> >
>> > When executed on the same train.arff and eval.arff, I obtained very
>> > different results for 2 different versions of weka SMO. Below are 
>> the
>> > commands:
>> >
>> > java -mx500m -cp /home/grad/chanys/sw/weka-3-4-5/weka.jar
>> > weka.classifiers.meta.MultiClassClassifier -t train.arff -T 
>> eval.arff
>> > -p 1 -W weka.classifiers.functions.SMO
>>  >
>> > java -mx500m -cp /home/grad/chanys/wsd/sw/weka-3-2
>> > weka.classifiers.MultiClassClassifier -t train.arff -T eval.arff -p 
>> 1
>> > -W weka.classifiers.SMO
>> >
>> > On the older version weka-3-2, the real-numbers returned for each 
>> test
>> > instance is averaging 0.7, but for the newer weka-3-4-5, the
>> > real-numbers returned for each test instance is 1.0 (i.e. 100%
>> > confident of belonging to a certain class!).
>> >
>> > What's the difference between these 2 versions? The reason why I 
>> want
>> > to switch to the newer version is because it can fit logistic models
>> > to the SVM outputs.
>> >
>> > Yee Seng._______________________________________________
>> > Wekalist mailing list
>> > Wekalist <at> list.scms.waikato.ac.nz
>> > https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
>>

headers

Natália Quirino de Oliveira | 2 Sep 2005 00:03

Tertius algorithm


  Hi guys, Im using the Tertius association rule algoritm, but I need some 
pseudocode or article about it. 

Same for Apriori. 

Thanks

_______________________________________________
Wekalist mailing list
Wekalist <at> list.scms.waikato.ac.nz
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist

headers

Peter Reutemann | 2 Sep 2005 00:13

Re: Tertius algorithm

>   Hi guys, Im using the Tertius association rule algoritm, but I need some 
> pseudocode or article about it. 
> 
> Same for Apriori. 

You'll find paper references in the Javadoc of each algorithm.

Cheers, Peter
--

-- 
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/     +64 (7) 838-4466 Ext. 5174

headers

Ashraf M. Kibriya | 2 Sep 2005 04:08

Re: StringToWordVector -W option

>From the GUI I don't think you can specify that option. However, if you
run the filter from the command line, there is an option "-c" that
allows you to set the class index. If you specify a class attribute
(index of the attribute) using this option then StringToWordVector will
try to keep top 100 words for class A, top 100 words for class B, and so on.

Regards,
Ashraf

Dwi Sianto Mansjur wrote:

> I have looked more into StringToWordVector
> I don't think there is an option to select top 100 words for class A ,
> top 100 words for class B, and so on
> If there is please let me know the option
>
> thanks
>
> On 8/23/05, *Ashraf M. Kibriya* <amk14 <at> cs.waikato.ac.nz
> <mailto:amk14 <at> cs.waikato.ac.nz>> wrote:
>
>     It keeps the 1000 highest occuring words, i.e. with the highest count
>     regardless of the number of documents it appears in.
>
>     It says in the documentation to be approaximate because if say for
>     example the 1000th word has frequency 20, and there are also another
>     lets say four words with frequency 20, then it will keep 1004 words,
>     including all those which have the same same count and are at the
>     boundary.
>
>
>     Regards,
>     Ashraf
>
>

Mon	Tue	Wed	Thu	Fri	Sat	Sun

			1	2	3	4
5	6	7	8	9	10	11
12	13	14	15	16	17	18
19	20	21	22	23	24	25
26	27	28	29	30

189	May 2013
358	April 2013
305	March 2013
302	February 2013
216	January 2013
200	December 2012
174	November 2012
178	October 2012
121	September 2012
180	August 2012
147	July 2012
192	June 2012
228	May 2012
220	April 2012
335	March 2012
233	February 2012
224	January 2012
285	December 2011
284	November 2011
247	October 2011
247	September 2011
326	August 2011
275	July 2011
249	June 2011
320	May 2011
274	April 2011
461	March 2011
365	February 2011
300	January 2011
234	December 2010
229	November 2010
338	October 2010
289	September 2010
349	August 2010
271	July 2010
345	June 2010
410	May 2010
336	April 2010
302	March 2010
264	February 2010
397	January 2010
332	December 2009
365	November 2009
362	October 2009
436	September 2009
453	August 2009
317	July 2009
455	June 2009
527	May 2009
501	April 2009
525	March 2009
362	February 2009
336	January 2009
222	December 2008
338	November 2008
331	October 2008
216	September 2008
250	August 2008
325	July 2008
282	June 2008
219	May 2008
354	April 2008
338	March 2008
264	February 2008
262	January 2008
197	December 2007
214	November 2007
288	October 2007
245	September 2007
176	August 2007
311	July 2007
276	June 2007
363	May 2007
410	April 2007
302	March 2007
231	February 2007
236	January 2007
155	December 2006
262	November 2006
197	October 2006
199	September 2006
188	August 2006
207	July 2006
289	June 2006
315	May 2006
289	April 2006
330	March 2006
238	February 2006
198	January 2006
212	December 2005
206	November 2005
253	October 2005
280	September 2005
271	August 2005
157	July 2005
267	June 2005
221	May 2005
286	April 2005
182	March 2005
117	February 2005
111	January 2005
108	December 2004
106	November 2004
81	October 2004
56	September 2004
69	August 2004
101	July 2004
143	June 2004
96	May 2004
87	April 2004
71	March 2004
104	February 2004
96	January 2004
70	December 2003
143	November 2003
97	October 2003
80	September 2003
87	August 2003
99	July 2003
90	June 2003
100	May 2003
123	April 2003
122	March 2003
119	February 2003
94	January 2003
40	December 2002
137	November 2002
122	October 2002
81	September 2002
55	August 2002
61	July 2002
43	June 2002
70	May 2002
35	April 2002
3	January 2001
2	March 2000