Re: The status in the Weka's GUI still "Building model in training data"

I'm using Naive Bayes classifier in the GUI.  There are  no error messages printed to
the log ?

2012/3/14 <wekalist-request@...>

Send Wekalist mailing list submissions to
       wekalist@...

To subscribe or unsubscribe via the World Wide Web, visit
       https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
or, via email, send a message with subject or body 'help' to
       wekalist-request@...

You can reach the person managing the list at
       wekalist-owner@...

When replying, please edit your Subject line so it is more specific
than "Re: Contents of Wekalist digest..."


Today's Topics:

  1. Estimating Generalization Error (Abhijeet Godase)
  2. predictions with clustering (nellams@...)
  3. Probability prediction on J48 (PiccoloBuddha)
  4. Re: A question about features combination (Nalleli Chuc)
  5. Re: Using trained models. (Sean McCauliff)
  6. Re: The status in the Weka's GUI still "Building model    in
     training data" (Mark Hall)
  7. Re: Different results for SMO but one executed under Weka's
     GUI and another under Eclipse (Mark Hall)
  8. Re: Classifier Test Times (Mark Hall)
  9. Re: Training and Testing (Mark Hall)


----------------------------------------------------------------------

Message: 1
Date: Tue, 13 Mar 2012 15:27:28 +0530
From: Abhijeet Godase <abhigodase@...>
Subject: [Wekalist] Estimating Generalization Error
To: "Weka machine learning workbench list."
       <wekalist@...>
Message-ID:
       <CACrPG1C_cWcUkRGgE%2BQ5fLBPqJ-_E-yfBoJiJ%2BBW%2BFLr31rW9g@...>
Content-Type: text/plain; charset="iso-8859-1"

hi all,

I have a doubt regarding Estimation of Generalization Error.
Is it the Error on Training Set..? (like Mean Squared Error etc)
what are the different ways of calculating Generalization Error..?


--- Abhijeet Godase.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://list.scms.waikato.ac.nz/pipermail/wekalist/attachments/20120313/bbac3dc7/attachment-0001.html

------------------------------

Message: 2
Date: Tue, 13 Mar 2012 11:10:49 +0100 (CET)
From: "nellams@..." <nellams@...>
Subject: [Wekalist] predictions with clustering
To: wekalist@...
Message-ID:
       <4331373.3233541331633449076.JavaMail.defaultUser@defaultHost>
Content-Type: text/plain; charset="utf-8"

Hi all,How can I predict values of some attributes using clustering model in Java code?Thanks in advance,Antonella
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://list.scms.waikato.ac.nz/pipermail/wekalist/attachments/20120313/659731d9/attachment-0001.html

------------------------------

Message: 3
Date: Tue, 13 Mar 2012 09:27:58 -0700 (PDT)
From: PiccoloBuddha <nicola.lazza@...>
Subject: [Wekalist] Probability prediction on J48
To: wekalist@...
Message-ID: <33495175.post@...>
Content-Type: text/plain; charset=us-ascii


Hi i'm trying to use J48 algorithm. When i set to "output predictions"
probabilities that appears are referred to the single instance or to the
leaf associated to the instance ? I mean: all the instances contained in a
single leaf have the same probability?
--
View this message in context: http://old.nabble.com/Probability-prediction-on-J48-tp33495175p33495175.html
Sent from the WEKA mailing list archive at Nabble.com.




------------------------------

Message: 4
Date: Tue, 13 Mar 2012 14:51:47 -0600
From: Nalleli Chuc <nallelicg@...>
Subject: Re: [Wekalist] A question about features combination
To: "Weka machine learning workbench list."
       <wekalist@...>
Message-ID:
       <CAPvuwY75jr0NCPHG1LEr05nJqZornqaE2NL33ZW8EmVYAueKoA@...>
Content-Type: text/plain; charset="iso-8859-1"

Please remove from the list.
Thanks for sharing your knowledge, have been helpful.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://list.scms.waikato.ac.nz/pipermail/wekalist/attachments/20120313/6d0171b8/attachment-0001.html

------------------------------

Message: 5
Date: Tue, 13 Mar 2012 14:14:23 -0700
From: Sean McCauliff <Sean.D.McCauliff@...>
Subject: Re: [Wekalist] Using trained models.
To: <wekalist@...>
Message-ID: <4F5FB8AF.8070708@...>
Content-Type: text/plain; charset="ISO-8859-1"; format=flowed

On 03/11/2012 07:27 PM, Sean McCauliff wrote:
> What is the correct procedure for using trained models?
>
> Initially I just want to understand the training instances that are not
> classified correctly in order to know what is different about these
> instances that makes them hard to classify.  I tried setting "output
> predictions" to true on the explorer's classification options. And then
> run the classifier with the same parameters.  This outputs a list of
> instance numbers, their actual class, their predicted class and if it
> was misclassified.  This seems great.  Except that the instance numbers
> repeat so I can't actually associate them with lines in the ARFF file.
>

Seems like I needed to RTFM.  Once I turn off cross validation the
instance numbers make sense.

Sean




------------------------------

Message: 6
Date: Wed, 14 Mar 2012 19:35:57 +1300
From: Mark Hall <mhall@...>
Subject: Re: [Wekalist] The status in the Weka's GUI still "Building
       model   in training data"
To: Weka machine learning workbench list.
       <wekalist@...>
Message-ID: <4F603C4D.5060100@...>
Content-Type: text/plain; charset="ISO-8859-1"; format=flowed

On 9/03/12 4:01 PM, Nounou Biatrice wrote:
> Hi all,
> I'm using Weka's GUI and I'm running a classifier. My problem is :
> My data set contains 3995 Instances and 18118  Attributes. When I run
> the classifier, the status in the Weka's GUI still "Building model in
> training data".
> I waited for  a long time the status to be  changed to "building a model
> for fold 1" but the status remains unchanged. And the same problem
> occurs in eclipse.
> When  I unchecked  the box "output model" In the classifier evaluation
> options, the problem is solved. For me, I must have  the box "output
> model" checked because I want to know  the time needed to build the model.
> Could anyone explain to me how to proceed?

Which classifier are you using? Are there any error messages printed to
the log or console?

Cheers,
Mark.



------------------------------

Message: 7
Date: Wed, 14 Mar 2012 19:40:13 +1300
From: Mark Hall <mhall@...>
Subject: Re: [Wekalist] Different results for SMO but one executed
       under Weka's    GUI and another under Eclipse
To: Weka machine learning workbench list.
       <wekalist@...>
Message-ID: <4F603D4D.1080500@...>
Content-Type: text/plain; charset="ISO-8859-1"; format=flowed

On 9/03/12 6:25 AM, Nounou Biatrice wrote:
> I Know that WekaDemo uses 10 folds cv. But I forgot to tell you that I
> have changed this line:
>
> // 3fold CV with seed=1
>      m_Evaluation = new Evaluation(filtered);
>      m_Evaluation.crossValidateModel(
>          m_Classifier, filtered, 3, m_Training.getRandomNumberGenerator(1));
>
> Under eclipse, I have 90.9091 % but under Weka's GUI, I have 91.9192 %
> as accuracy.

The data is being randomized differently - here is the javadoc for
Instances.getRandomNumberGenerator():

/**
   * Returns a random number generator. The initial seed of the random
   * number generator depends on the given seed and the hash code of
   * a string representation of a instances chosen based on the given
   * seed.
   *
   * @param seed the given seed
   * @return the random number generator
   */

Cheers,
Mark.



------------------------------

Message: 8
Date: Wed, 14 Mar 2012 19:45:44 +1300
From: Mark Hall <mhall@...>
Subject: Re: [Wekalist] Classifier Test Times
To: Nandita Tripathi <tnandita@...>
Cc: "Weka machine learning workbench list."
       <wekalist@...>
Message-ID: <4F603E98.3030804@...>
Content-Type: text/plain; charset="ISO-8859-1"; format=flowed

On 14/03/12 7:43 PM, Nandita Tripathi wrote:
>
>
> Hi
>
> I am working with the experimenter. Does this mean that test time is the average test time per test set for the experimenter - not the average test time per test vector.

The experimenter reports averages over folds/runs for all the metrics it
computes.

Cheers,
Mark.



------------------------------

Message: 9
Date: Wed, 14 Mar 2012 19:50:41 +1300
From: Mark Hall <mhall@...>
Subject: Re: [Wekalist] Training and Testing
To: Weka machine learning workbench list.
       <wekalist@...>
Message-ID: <4F603FC1.1000907@...>
Content-Type: text/plain; charset="ISO-8859-1"; format=flowed

On 10/03/12 3:37 AM, Sonal Kothari wrote:
> I have had the same problem. Would using the input mapped classifier
> help in such a case ?

It will help, and should allow your experiments to run without errors.
You should be aware that if there are some labels that occur in test CSV
files and not in training (and vice versa) then results might be
slightly different than if ARFF files with unified headers are created.
This is because nominal values that occur in test sets but not in
training sets result in a missing value being set for that attribute in
the test instance before it is passed to the classifier for prediction.

Cheers,
Mark.



------------------------------

_______________________________________________
Wekalist mailing list
Wekalist@...
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist


End of Wekalist Digest, Vol 109, Issue 28
*****************************************

--

Best regards,

---

"Someone is sitting in the shade today, because someone planted a tree a long time ago." Warren Buffett.

---