Nabble - WEKA

20+ classification algorithms

2009-11-09T20:56:46Z

I have a list of algorithms and I was wondering how many of them are already in weka. If anyone could let me know which ones are which, sometimes I know the names can be deceiving. Or a good source that gives descriptions of the algorithms weka has.

The list:
FACT, univariate
FACT, linear
C4.5 trees
C4.5 rules
OC1, univariate
OC1, linear
OC1, mixed
S-PLUS tree, 0-SE
S-PLUS tree, 1-SE
LMDT, linear
CAL5
T1, single split
Linear discriminant analysis
Quadratic discriminant analysis
Nearest-neighbor
Linear logistic regression
Flexible discriminant analysis, degree 1
FDA, degree 2
Penalized LDA
Mixture discriminant analysis
POLYCLASS
Learning vector quantization
Radial basis function network

Any help would be much appreciated

Re: AdaBoost.M1: How base learner labels an instance

2009-11-09T16:16:16Z

On Mon, Nov 9, 2009 at 6:10 PM, Peter Reutemann <fracpete@...> wrote:

"distributionForInstance" returns a double array, one element for each
label of the class attribute (containing the probability for this
particular label). "classifyInstance" returns the chosen index among
the labels, i.e., the one with the highest probability.

This is also covered in the wiki article "Use Weka in your Java code",
section "Classifying instances ":
http://weka.wikispaces.com/Use+Weka+in+your+Java+code

If you want to compare the class labels, actual and predicted, just
use the value that "classifyInstance" returns (= predicted) and the
one that "Instance.classValue()" returns (= actual). No real
post-processing necessary.

Thanks for your quick reply, Peter!

This disuccsion guides me further into the implementation of weight udpating. Instead of decreasing the weights of correctly labeled instances and increasing those incorrectly labeled as described in the original paper of Freund and Schapire, weka implements AdaBoost.M1 to keep the weights unchanged for the correctly labeled instances and increase those incorrectly labeled, as shown in setWeights() of AdaBoost.M1 (below). Different ways to implement the same logic.

******************************************************************************

protected void setWeights(Instances training, double reweight)
throws Exception {

double oldSumOfWeights, newSumOfWeights;

    oldSumOfWeights = training.sumOfWeights();
    Enumeration enu = training.enumerateInstances();
    while (enu.hasMoreElements()) {
      Instance instance = (Instance) enu.nextElement();
      if (!Utils.eq(m_Classifiers[m_NumIterationsPerformed].classifyInstance(instance),
      instance.classValue()))
       instance.setWeight(instance.weight() * reweight);
    }
    .....

********************************************************************

Best,

bbl

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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: question on saving the results buffer in knowledge flow

2009-11-09T15:32:30Z

> Today I created the attached file prediction_creator_evaluation.kfml .
> Using the attached train and test set I want to evaluate and save the
> results I visualize by using a text viewer component of the knowledge flow.
> Unfortunatly, I cannot save the produced results automatically, manually it
> is possible by a doing a righter mouse click and choosing save result
> buffer. Is it possible to save these results automatically?

No, you can't save them automatically, as far as I know.

Have you thought about using the commandline?

java weka.classifiers.bayes.NaiveBayes -t train.arff -T test.arff > out.txt

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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: remove dubble string after the use of a stemming algorithm

2009-11-09T15:26:49Z

> In weka this method works fine and the result is
> good. However, I cannot apply the same method to the file I attached. For
> some reason it does not use the -R option to only select the first
> attribute. Can you help?

The filter automatically only works on STRING attributes. If the
dataset contains only a single STRING attribute, like in your case,
then you can just leave "first-last". The more important thing is,
what did you use as class attribute? In your case, the last attribute
is a numeric one. The filter behaves differently if you don't define a
class attribute, choose a numeric one or a nominal one. By default,
the filter operates on a per class (label) basis.

I had no problem processing your dataset, using a post-3.7.0 release
code basis. What's the exact problem that you're experiencing?

BTW please don't top-post (does anybody bother reading the mailing
list etiquette??).

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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: AdaBoost.M1: How base learner labels an instance

2009-11-09T15:10:03Z

> When doing some binary experiments with AdaBoost.M1 (DecisionStump as base
> learner), I got a question and want to confirm on how a decisionstump labels
> an instance.
>
>
> When Evaluator class evaluates a decisionstump on an instance, it calls the
> method below and returns a double value pred as the instance's label. pred
> is the index of the largest value of the array dist[] that is the class
> probabilities of the current instance. For example, with the dist[] array
> {0.61, 0.39} for an instance on the current decisionstump, pred is returned
> with value 0 since the 0th value of dist[] (0.61) is larger than its 1st
> value(0.39).
>
> My consufusion is that, with the class labels {-1, 1} in my ARFF file, I
> would assume the decisionstump gives me either -1 or 1 as the label, so that
> I can direclty tell which instance is incorrectly labeled. However, what I
> get is either 0 or 1. Thus, I need to write some post-processing code to
> interpret the class label assignments. That is, if dist[0] aligns with class
> label -1, and dist[1] aligns with class label 1, pred value 0 means the
> current instance is labeled as -1, and pred value 1 means the current
> instance is labeled as 1.

"distributionForInstance" returns a double array, one element for each
label of the class attribute (containing the probability for this
particular label). "classifyInstance" returns the chosen index among
the labels, i.e., the one with the highest probability.

This is also covered in the wiki article "Use Weka in your Java code",
section "Classifying instances ":
http://weka.wikispaces.com/Use+Weka+in+your+Java+code

If you want to compare the class labels, actual and predicted, just
use the value that "classifyInstance" returns (= predicted) and the
one that "Instance.classValue()" returns (= actual). No real
post-processing necessary.

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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

AdaBoost.M1: How base learner labels an instance

2009-11-09T14:22:49Z

Hi,

When doing some binary experiments with AdaBoost.M1 (DecisionStump as base learner), I got a question and want to confirm on how a decisionstump labels an instance.

When Evaluator class evaluates a decisionstump on an instance, it calls the method below and returns a double value pred as the instance's label. pred is the index of the largest value of the array dist[] that is the class probabilities of the current instance. For example, with the dist[] array {0.61, 0.39} for an instance on the current decisionstump, pred is returned with value 0 since the 0th value of dist[] (0.61) is larger than its 1st value(0.39).

My consufusion is that, with the class labels {-1, 1} in my ARFF file, I would assume the decisionstump gives me either -1 or 1 as the label, so that I can direclty tell which instance is incorrectly labeled. However, what I get is either 0 or 1. Thus, I need to write some post-processing code to interpret the class label assignments. That is, if dist[0] aligns with class label -1, and dist[1] aligns with class label 1, pred value 0 means the current instance is labeled as -1, and pred value 1 means the current instance is labeled as 1.

**********************************************************************************************************
public double evaluateModelOnceAndRecordPrediction(Classifier classifier,
      Instance instance) throws Exception {

    Instance classMissing = (Instance)instance.copy();
    double pred =
    classMissing.setDataset(instance.dataset());
    classMissing.setClassMissing();
    if (m_ClassIsNominal) {
      if (m_Predictions == null) {
    m_Predictions = new FastVector();
      }
      double [] dist = classifier.distributionForInstance(classMissing);
      pred = Utils.maxIndex(dist);
      if (dist[(int)pred] <= 0) {
    pred = Instance.missingValue();
      }
      updateStatsForClassifier(dist, instance);
      m_Predictions.addElement(new NominalPrediction(instance.classValue(), dist,
    instance.weight()));
    } else {
      pred = classifier.classifyInstance(classMissing);
      updateStatsForPredictor(pred, instance);
    }
    return pred;
}
********************************************************************************************

Any comments? Thanks.....

bbl

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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: Cost sensitive Evaluation

2009-11-09T11:18:33Z

All Good wrote:
> Its trouble.
>
> LinearRegression does not work with Nominal Class
> and cost Sensitive evaluations wants us to have Nominal class.

Try wrapping LinearRegression in
weka.classifiers.meta.ClassificationViaRegression.

Cheers,
Mark.

--
Mark Hall
Senior Developer/Consultant, Pentaho Open Source Business Intelligence
Citadel International, Suite 340, 5950 Hazeltine National Dr., Orlando,
FL 32822, USA
+64 7 348-7099 office, +64 21 399-132 mobile, +1 815 550-8637 fax,
Skype: mark.andrew.hall, Yahoo: mark_andrew_hall
Download the latest release today
<http://www.sourceforge.net/projects/pentaho>

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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 Linear Regression Problem

2009-11-09T11:11:45Z

Please no top-posting...

> One of the problem is that I have quite a big difference between Positive
> and Negative Class levels.
>
> Non Relevant instances are much more than Relevant instance (NR are more
> than double of RV instances). So I think I should use COST Sensitive
> Evaluation. I have read something here
> http://wekadocs.com/node/15
>
> but still cannot understand that how should I set this Cost Sensitive
> matrix. Bz in ths matrix shown on this web, I dont know what value
> corresponds to what label ...
>
> % Rows Columns
>
> 2 2
>
> % Matrix elements
>
> 0 2
>
> 1 0
>
> like in above matrix, Where I need to put 2 (to tell the classifier that
> misclassifying a RELEVANT instance is Twice as Expensive than misclassifying
> the NR one).

The maxtrix is just the costs for the confusion matrix. On the
diagonal, you have the correctly classified ones and all other cells
are for misclassified ones.

Here is an example of a confusion matrix (J48 on the UCI dataset "iris"):

a b c <-- classified as
49 1 0 | a = Iris-setosa
0 47 3 | b = Iris-versicolor
0 2 48 | c = Iris-virginica

The first row contains all the instances that have label "Iris-setosa"
(= a). 1 instance got misclassified as "Iris-versicolor" (= b).
The cost matrix merely places "costs" in the misclassification cells.

See also the following wiki article:
http://weka.wikispaces.com/CostSensitiveClassifier

> Also if you can inform me that what class is for LogisticRegression and
> RegressionSVM in Weka??

weka.classifiers.functions.Logistic
weka.classifiers.functions.SVMreg or SMOreg (depends on your version of Weka)

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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: Cost sensitive Evaluation

2009-11-09T10:36:32Z

> LinearRegression does not work with Nominal Class
> and cost Sensitive evaluations wants us to have Nominal class.
>
> What is the solution :(

Like I already posted earlier: you need to discretize your numeric
class, using Weka's unsupervised Discretize filter (package
weka.filters.unsupervised.attribute).

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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: Inherent structure within documents

2009-11-09T09:28:10Z

If you want to find the inherent structure in just one document, break it up into separate documents that are likely to exhibit the same structure. Weka will be able to suggest correlations between word usage; clustering can suggest topics.

Weka can't find out what Chapter 1 and Chapter 2 are. That would be trying to find differences in text, and in text analysis, all you have are differences.

You could take all the paragraphs out of one chapter and make them each their own document, then cluster that, or take all paragraphs out of all chapters and make them their own document with the class of the chapter they're in, and classify that.

On Sun, Nov 8, 2009 at 10:49 PM, Vasundhara Chaudhuri Chakraborty <vasuchau@...> wrote:

Hi,

I am new to weka and was wondering if someone could help me with this.

I have some documents and it is expected that these documents have some inherent structure. So I want to use clustring techniques and see what kind of structure can be found.

However I am not trying to cluster the documents rather I am trying to find the inherent structure within documents, if any.

Can this be done using weka?

thanks a lot for your help.

sincerely,

VC

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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: remove dubble string after the use of a stemming algorithm

2009-11-09T07:50:03Z

Hi Peter,

Thank you for the response. In weka this method works fine and the result is
good. However, I cannot apply the same method to the file I attached. For
some reason it does not use the -R option to only select the first
attribute. Can you help?

Kind regards,

Paul

-----Oorspronkelijk bericht-----
Van: wekalist-bounces@...
[mailto:wekalist-bounces@...] Namens Peter Reutemann
Verzonden: maandag 9 november 2009 8:57
Aan: Weka machine learning workbench list.
Onderwerp: Re: [Wekalist] remove dubble string after the use of a stemming
algorithm

> At the moment I am using the dutch wordstemmer separately. When
> applying the stemmer to a dataset with a large amount of single words,
> the output of the stringtowordvector filter, e.g. the attached file,
> the output has the same amount of features. Is it possible to reduce
> the number of unique features and adapting the frequencies of the stemmed
words?

>
> For instance:
>
> Nederlandse
> Nederlands
> Nederlanders
> Nederlander
> Nederland
>
> Becomes
>
> Nederland
> Nederland
> Nederlander
> Nederlander
> Nederland
>
> And should become
>
> Nederland (freq 3)
> Nederlander (freq 2)

Instead of stemming the text outside Weka, I've converted your data into an
ARFF file (see attached file) and perform all preprocessing within Weka
using the StringToWordVector filter.

As stemmer I'm using the "dutch" stemmer of the snowball stemmers (see FAQ
"The snowball stemmers don't work, what am I doing wrong?", if you haven't
added them to Weka yet). When not using word counts, then the generated
"Nederland" and "Nederlander" attributes contain only 1 (= binary attribute,
word occurs or not). But if I turn on word counts (outputWordCounts/-C),
then I get 3 for "Nederland" and 2 for "Nederlander".

Does that help?

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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

feature_selection_nominal_3696.arff (188K) Download Attachment

Cost sensitive Evaluation

2009-11-09T05:47:14Z

Its trouble.

LinearRegression does not work with Nominal Class
and cost Sensitive evaluations wants us to have Nominal class.

What is the solution :(

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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 Linear Regression Problem

2009-11-09T03:59:06Z

One of the problem is that I have quite a big difference between Positive and Negative Class levels.

Non Relevant instances are much more than Relevant instance (NR are more than double of RV instances). So I think I should use COST Sensitive Evaluation. I have read something here
http://wekadocs.com/node/15

but still cannot understand that how should I set this Cost Sensitive matrix. Bz in ths matrix shown on this web, I dont know what value corresponds to what label ...

% Rows Columns

2 2

% Matrix elements

0 2

1 0

like in above matrix, Where I need to put 2 (to tell the classifier that misclassifying a RELEVANT instance is Twice as Expensive than misclassifying the NR one).

Also if you can inform me that what class is for LogisticRegression and RegressionSVM in Weka??

thanks

On Mon, Nov 9, 2009 at 10:36 AM, All Good <kimskams80@...> wrote:

yes, thanks dear .. I just read it ....

But I think you suggested this just because you thought I have a nominal class attribute.. So it might give same results .. as Linear Regression if they are both same. I will try using Cross Validation and let see the results .. but I am really thankful to you for all your kind help.

Thanks a Lot and god bless you

On Mon, Nov 9, 2009 at 10:30 AM, Peter Reutemann <fracpete@...> wrote:

> I thought you're using LinearRegression, but your dataset seems to
> have a binary class attribute instead of numeric one.
>>>> I am using 1 for Relevant and 0 for Non-Relevant
> P@5 means we rank all results by the predictions GIVEN by Regression and
> then rank them by those values ... For Top 5 , we look at their actual CLASS
> and then calculate Precision @5. I will read about MetaClassifier bz with
> current Setup of Training Data, its Line in not enabled in Explorer may be
> it requires data in different format.

The meta-classifier needs a nominal class attribute. It doesn't work
with numeric class attributes. You'd have to discretize the class
attribute first (using Weka's unsupervised Discretize filter, for
instance).

[...]

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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

CFP: ISNN2010 (June 6-10, 2010; Shanghai, China)

2009-11-09T03:21:47Z

CALL FOR PAPERS

7th International Symposium on Neural Networks (ISNN2010)
June 6-10, 2010; Shanghai, China

Websites: http://isnn2010.sjtu.edu.cn or http://isnn2010.mae.cuhk.edu.hk

Paper submissions: http://isnn2010.sjtu.edu.cn/login.asp

The Seventh International Symposium on Neural Networks (ISNN 2010) will be held in Shanghai, following the successes of previous events. Shanghai is the largest city in China, located in her eastern coast at the outlet of the Yangtze River. Originally a fishing and textiles town, Shanghai grew to importance in the 19th century. In 2005 Shanghai became the world's busiest cargo port. The city is an emerging tourist destination renowned for its historical landmarks such as the Bund and Xintiandi, its modern and ever-expanding Pudong skyline including the Oriental Pearl Tower, and its new reputation as a cosmopolitan center of culture and design. Today, Shanghai is the largest center of commerce and finance in mainland China, and has been described as the “showpiece” of the world's fastest-growing economy. In addition, Shanghai is the venue of forthcoming World Expo 2010 to take place from May 1 to October 31 (the symposium registration includes one day tour to the World Expo on June 10).

ISNN 2010 aims to provide a high-level international forum for scientists, engineers, and educators to present the state of the art of neural network research and applications in related fields. The symposium will feature plenary speeches given by world renowned scholars, regular sessions with broad coverage, and special sessions focusing on popular topics.

Prospective authors are invited to contribute high-quality papers to ISNN 2010. In addition, proposals for special sessions within the technical scopes of the symposium are solicited. Special sessions, to be organized by internationally recognized experts, aim to bring together researchers in special focused topics. Papers submitted for special sessions are to be peer-reviewed with the same criteria used for the contributed papers. Researchers interested in organizing special sessions are invited to submit formal proposals to ISNN 2010. A special session proposal should include the session title, a brief description of the scope and motivation, names, contact information and brief biographical information on the organizers.

Authors are invited to submit full-length papers (10 pages maximum) by the submission deadline through the online submission system. Potential organizers are also invited to enlist five or more papers with cohesive topics to form special sessions. The submission of a paper implies that the paper is original and has not been submitted under review or copyright-protected elsewhere and will be presented by an author if accepted. All submitted papers will be refereed by experts in the field based on the criteria of originality, significance, quality, and clarity. The authors of accepted papers will have an opportunity to revise their papers and take consideration of the referees' comments and suggestions. The ISNN 2010 proceedings will be published by Springer in its series of Lecture Notes in Computer Science (EI) and Advances in Intelligent and Soft Computing (ISTP). Selected good papers will be included in special issues of several journals such as Neurocomputing, Neural Computation and Applications, Cognitive Neurodynamics, Cognitive Computation, and Mathematics and Computers in Simulation. In addition, the International Neural Network Society (INNS) will offer two best student paper awards (US$250 each with one-year INNS membership).

************************************************
TOPIC AREAS
************************************************

1. Computational Neuroscience and Cognitive Science
Spiking neurons
Visual and auditory cortex
Neural encoding and decoding
Plasticity and adaptation
Brain imaging
Learning and memory
Inference and reasoning
Perception, emotion and development
Attractor and associative memory
Neurodynamics and complex systems

2. Models, Methods and Inference
Stability and convergence analysis
Neural network models
Supervised learning
Unsupervised learning
Embeddings and manifold learning
Active learning
Statistical and informationtheoretic methods
Kernel methods and support vector machines
Mixture models
Graphical and causal models
Bayesian networks
Topic models
Gaussian processes
Model selection
Matrix/tensor analysis
Structured and relational data
Clustering

3. Vision and Auditory Modelling
Visual perception and modelling
Visual selective attention
Statistical and pattern recognition
Visual coding and representation
Object recognition
Motion and tracking
Natural scene analysis
Auditory perception and modelling
Source separation
Speech recognition and speech synthesis
Speaker identification
Audio and speech retrieval
Music modelling and analysis

4. Control, Robotics and Hardware
Neuromorphic hardware and implementation
Embedded neural networks
Fuzzy neural networks
Cognitive robotics
Developmental robotics
Multi-agent systems and game theory
Reinforcement learning
Planning and decision making
Action and motor control
Visuomotor control

5. Novel Approaches and Applications
Brain-like systems
Adaptive intelligent systems
Brain-computer interfaces
Granular computing
Hybrid intelligent systems
Neuroinformatics and neuroengineering
Bioinformatics
Information retrieval
Data mining and knowledge discovery
Natural language processing

************************************************
IMPORTANT DATES
************************************************

Full paper submission deadline: December 1, 2009
Notification of acceptance: January 1, 2010
Camera-ready copy and author registration: February 1, 2010

For inquiries, please contact the secretariat at isnn2010@sjtu.edu.cn

Re: A Linear Regression Problem

2009-11-09T01:36:50Z

yes, thanks dear .. I just read it ....

But I think you suggested this just because you thought I have a nominal class attribute.. So it might give same results .. as Linear Regression if they are both same. I will try using Cross Validation and let see the results .. but I am really thankful to you for all your kind help.

Thanks a Lot and god bless you

On Mon, Nov 9, 2009 at 10:30 AM, Peter Reutemann <fracpete@...> wrote:

> I thought you're using LinearRegression, but your dataset seems to
> have a binary class attribute instead of numeric one.
>>>> I am using 1 for Relevant and 0 for Non-Relevant
> P@5 means we rank all results by the predictions GIVEN by Regression and
> then rank them by those values ... For Top 5 , we look at their actual CLASS
> and then calculate Precision @5. I will read about MetaClassifier bz with
> current Setup of Training Data, its Line in not enabled in Explorer may be
> it requires data in different format.

The meta-classifier needs a nominal class attribute. It doesn't work
with numeric class attributes. You'd have to discretize the class
attribute first (using Weka's unsupervised Discretize filter, for
instance).

[...]

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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 Linear Regression Problem

2009-11-09T01:30:50Z

> I thought you're using LinearRegression, but your dataset seems to
> have a binary class attribute instead of numeric one.
>>>> I am using 1 for Relevant and 0 for Non-Relevant
> P@5 means we rank all results by the predictions GIVEN by Regression and
> then rank them by those values ... For Top 5 , we look at their actual CLASS
> and then calculate Precision @5. I will read about MetaClassifier bz with
> current Setup of Training Data, its Line in not enabled in Explorer may be
> it requires data in different format.

The meta-classifier needs a nominal class attribute. It doesn't work
with numeric class attributes. You'd have to discretize the class
attribute first (using Weka's unsupervised Discretize filter, for
instance).

[...]

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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 Linear Regression Problem

2009-11-09T01:25:17Z

I thought you're using LinearRegression, but your dataset seems to
have a binary class attribute instead of numeric one.
>>> I am using 1 for Relevant and 0 for Non-Relevant
P@5 means we rank all results by the predictions GIVEN by Regression and then rank them by those values ... For Top 5 , we look at their actual CLASS and then calculate Precision @5. I will read about MetaClassifier bz with current Setup of Training Data, its Line in not enabled in Explorer may be it requires data in different format.

On Mon, Nov 9, 2009 at 10:15 AM, Peter Reutemann <fracpete@...> wrote:

Please no top-posting, see mailing list etiquette why
(http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html).

> Its not samples or instances. You can call it Queries for them we need to
> find the RELEVANT Documents. So in Training Data (using 10 Queries), we have
> almost 7800 instances and for Testing data we have almost 8330 instances and
> on confirmation data (5 Queries) we have 5150 instances (i.e. labels of
> class is Relevant or NonRelevant for each instance).

> So its quite a lot I think.

it's reasonable large, yes.

> As I explained we are checking on 3 different
> combinations of Features (using one fetaure in comb1, 8 features in Comb2
> and 10 features in comb3). Theoretically the results (Precision@5 being
> calculated by results of Regression by our program) of regression should be
> like

Not sure what you mean with "Precision@5".

> Comb1<Comb2<Comb3
>
> It gives same results for Conofirmation Data but on Testing data one
> exception occurs where result of Comb1> Comb2
>
> So its kind of strange.
>
> (The procedure I am using while using Linear Regression is .... that I give
> the Training File to Regression and then give it Testibg Data which gives
> some results and then our program use those results to have results P@5 on
> Testing Data ... Similarly I use the Confirmation Data to have its
> results..... )
>
> May be I am doing some mistake somewhere if you can mention ...

Since you seem to perform some kind of post-processing of the
regression output, it's hard to figure what's going wrong. One reason
could be, that the choice of datasets (train, test, validation) is
unfortunate, resulting in this unexpected behavior that you described.

Have you thought about using the meta-classifier
ClassificationViaRegression, in case your dataset has a binary class
attribute? Then you could perform cross-validation (and in the
Experimenter even multiple runs of cross-validation) in order to
obtain more reliable statistics.

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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 Linear Regression Problem

2009-11-09T01:15:52Z

Please no top-posting, see mailing list etiquette why
(http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html).

> Its not samples or instances. You can call it Queries for them we need to
> find the RELEVANT Documents. So in Training Data (using 10 Queries), we have
> almost 7800 instances and for Testing data we have almost 8330 instances and
> on confirmation data (5 Queries) we have 5150 instances (i.e. labels of
> class is Relevant or NonRelevant for each instance).

I thought you're using LinearRegression, but your dataset seems to
have a binary class attribute instead of numeric one.

> So its quite a lot I think.

it's reasonable large, yes.

> As I explained we are checking on 3 different
> combinations of Features (using one fetaure in comb1, 8 features in Comb2
> and 10 features in comb3). Theoretically the results (Precision@5 being
> calculated by results of Regression by our program) of regression should be
> like

Not sure what you mean with "Precision@5".

> Comb1<Comb2<Comb3
>
> It gives same results for Conofirmation Data but on Testing data one
> exception occurs where result of Comb1> Comb2
>
> So its kind of strange.
>
> (The procedure I am using while using Linear Regression is .... that I give
> the Training File to Regression and then give it Testibg Data which gives
> some results and then our program use those results to have results P@5 on
> Testing Data ... Similarly I use the Confirmation Data to have its
> results..... )
>
> May be I am doing some mistake somewhere if you can mention ...

Since you seem to perform some kind of post-processing of the
regression output, it's hard to figure what's going wrong. One reason
could be, that the choice of datasets (train, test, validation) is
unfortunate, resulting in this unexpected behavior that you described.

Have you thought about using the meta-classifier
ClassificationViaRegression, in case your dataset has a binary class
attribute? Then you could perform cross-validation (and in the
Experimenter even multiple runs of cross-validation) in order to
obtain more reliable statistics.

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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 Linear Regression Problem

2009-11-09T00:56:33Z

Sorry that I did not explained in detail. Let me write in details:

Its not samples or instances. You can call it Queries for them we need to find the RELEVANT Documents. So in Training Data (using 10 Queries), we have almost 7800 instances and for Testing data we have almost 8330 instances and on confirmation data (5 Queries) we have 5150 instances (i.e. labels of class is Relevant or NonRelevant for each instance).
So its quite a lot I think. As I explained we are checking on 3 different combinations of Features (using one fetaure in comb1, 8 features in Comb2 and 10 features in comb3). Theoretically the results (Precision@5 being calculated by results of Regression by our program) of regression should be like

Comb1<Comb2<Comb3

It gives same results for Conofirmation Data but on Testing data one exception occurs where result of Comb1> Comb2

So its kind of strange.

(The procedure I am using while using Linear Regression is .... that I give the Training File to Regression and then give it Testibg Data which gives some results and then our program use those results to have results P@5 on Testing Data ... Similarly I use the Confirmation Data to have its results..... )

May be I am doing some mistake somewhere if you can mention ...

Thanks in Advance

On Mon, Nov 9, 2009 at 8:36 AM, Peter Reutemann <fracpete@...> wrote:

> We have a set of 25 Topics. We use 10 Topics for training and 10 for Testing
> and we kept 5 for confirmation of the results with 3 DIFERENT Combinations
> of FEATURES (so that we can show what features are more important).
>
> The results are same as we expecting except one strange exception is
> occuring. As we expecting, the result of Comb2 should be larger than Comb1
> which is demonstrated when testing on 5 Topics (for confirmation) but its
> less on TESTING DATA (ie Comb2 < Comb1) .
>
> Do you think its problem of Less Training Data or what??

I'm not quite sure what you mean with 25 topics. Are these 25 data
samples, i.e., rows/instances in the dataset? If so, 25 samples are
very little data... I personally prefer working with thousands of
examples.

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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: remove dubble string after the use of a stemming algorithm

2009-11-08T23:57:01Z

> At the moment I am using the dutch wordstemmer separately. When applying the

> stemmer to a dataset with a large amount of single words, the output of the
> stringtowordvector filter, e.g. the attached file, the output has the same
> amount of features. Is it possible to reduce the number of unique features
> and adapting the frequencies of the stemmed words?
>
> For instance:
>
> Nederlandse
> Nederlands
> Nederlanders
> Nederlander
> Nederland
>
> Becomes
>
> Nederland
> Nederland
> Nederlander
> Nederlander
> Nederland
>
> And should become
>
> Nederland (freq 3)
> Nederlander (freq 2)

Instead of stemming the text outside Weka, I've converted your data
into an ARFF file (see attached file) and perform all preprocessing
within Weka using the StringToWordVector filter.

As stemmer I'm using the "dutch" stemmer of the snowball stemmers (see
FAQ "The snowball stemmers don't work, what am I doing wrong?", if you
haven't added them to Weka yet). When not using word counts, then the
generated "Nederland" and "Nederlander" attributes contain only 1 (=
binary attribute, word occurs or not). But if I turn on word counts
(outputWordCounts/-C), then I get 3 for "Nederland" and 2 for
"Nederlander".

Does that help?

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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

dutch.arff (192 bytes) Download Attachment

Re: A Linear Regression Problem

2009-11-08T23:36:05Z

> We have a set of 25 Topics. We use 10 Topics for training and 10 for Testing
> and we kept 5 for confirmation of the results with 3 DIFERENT Combinations
> of FEATURES (so that we can show what features are more important).
>
> The results are same as we expecting except one strange exception is
> occuring. As we expecting, the result of Comb2 should be larger than Comb1
> which is demonstrated when testing on 5 Topics (for confirmation) but its
> less on TESTING DATA (ie Comb2 < Comb1) .
>
> Do you think its problem of Less Training Data or what??

I'm not quite sure what you mean with 25 topics. Are these 25 data
samples, i.e., rows/instances in the dataset? If so, 25 samples are
very little data... I personally prefer working with thousands of
examples.

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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: Inherent structure within documents

2009-11-08T23:27:45Z

> I have some documents and it is expected that these documents have some
> inherent structure. So I want to use clustring techniques and see what kind
> of structure can be found.
> However I am not trying to cluster the documents rather I am trying to find
> the inherent structure within documents, if any.
> Can this be done using weka?

Weka's clustering algorithms don't work on structures. Text mining in
Weka always involves turning the documents (i.e., plain text strings)
into numeric attributes (word occurrences, word counts, tf/idf, etc)
before applying any algorithm. Unless your preprocessing step involves
storing information about the structure in the documents itself
(something you'd have to implement yourself), the structural
information will get lost in the process.

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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

Inherent structure within documents

2009-11-08T19:49:53Z

Hi,

I am new to weka and was wondering if someone could help me with this.

I have some documents and it is expected that these documents have some inherent structure. So I want to use clustring techniques and see what kind of structure can be found.

However I am not trying to cluster the documents rather I am trying to find the inherent structure within documents, if any.

Can this be done using weka?

thanks a lot for your help.

sincerely,

Re: RE: Wekalist Digest, Vol 81, Issue 12

2009-11-08T11:09:58Z

Please don't reply to message digests (one gets enough spam nowadays
already), but start a new email with an appropriate subject instead.
Nobody's interested in getting hundreds of irrelevant lines of content
from the digest and a meagre 2 lines of actual question (and you
shouldn't top-post anyway, according to the mailing list etiquette!).

> Does anyone know whether Weka support LSA or not?

Yes, Weka supports LSA (only >= 3.6.1 and >= 3.7.0):
weka.attributeSelection.LatentSemanticAnalysis

> If it does, how can I use it?

>From Java or general usage via the Explorer?
- Java
see wiki article "Use Weka in your Java code", section "Attribute
selection" (referenced in FAQ "How do I use WEKA's classes in my own
code?")
- general usage
see FAQ "How do I perform attribute selection?"

Link to the FAQs available from the Weka homepage.

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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 ensemble

2009-11-08T10:59:25Z

> I am working on an ensemble algorithm to combine classifiers to improve
> the classification accuracy. I found that many Weka's classes such as
> Boosting, Bagging and etc. which seem quite relevant to me. Besides that,
> I think I've seen people around writing java code to make use of these
> classes. Does anybody knows or can point me locations that can run the
> algorithms from java code? Is the same way as in
> http://weka.wikispaces.com/Programmatic+Use can be used to run weka's
> ensemble algorithms?

All classifiers are derived from the same superclass,
weka.classifiers.Classifier (or if you work off subversion on the
developer version, then this is an interface now), hence they all work
the same. Depending on what kind of ensemble you're trying to
implement, you'll have to pick an appropriate superclass from the
weka.classifiers package (e.g., randomizable, with several base
classifier -> RandomizableMultipleClassifiersCombiner). See FAQ "How
do I write a new classifier or filter?" for more information (link to
FAQs available from the Weka homepage).

You might also want to take a look at the article "Use WEKA in your
Java code", which is a bit more elaborate than then "Programmatic Use"
one:
http://weka.wikispaces.com/Use+WEKA+in+your+Java+code

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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

AIPR-10 Call for papers

2009-11-08T09:21:43Z

AIPR-10 Call for papers

The 2010 International Conference on Artificial Intelligence and Pattern Recognition (AIPR-10) will be held during 12-14 of July 2010 in Orlando, FL, USA. AIPR is an important event in the areas of Artificial Intelligence (AI) as well as Pattern Recognition (PR) and focuses on all areas of AI, PR and related topics. The conference will be held at the same time and location where several other major international conferences will be taking place. The conference will be held as part of 2010 multi-conference (MULTICONF-10).

MULTICONF-10 (website: http://www.promoteresearch.org) will be held during July 12-14, 2010 in Orlando, Florida, USA. The primary goal of MULTICONF is to promote research and developmental activities in computer science, information technology, control engineering, and related fields. Another goal is to promote the dissemination of research to a multidisciplinary audience and to facilitate communication among researchers, developers, practitioners in different fields. The following conferences are planned to be organized as part of MULTICONF-10.

International Conference on Artificial Intelligence and Pattern Recognition (AIPR-10)
International Conference on Automation, Robotics and Control Systems (ARCS-10)
International Conference on Bioinformatics, Computational Biology, Genomics and Chemoinformatics (BCBGC-10)
International Conference on Computer Networks (CN-10)
International Conference on Enterprise Information Systems and Web Technologies (EISWT-10)
International Conference on High Performance Computing Systems (HPCS-10)
International Conference on Information Security and Privacy (ISP-10)
International Conference on Image and Video Processing and Computer Vision (IVPCV-10)
International Conference on Software Engineering Theory and Practice (SETP-10)
International Conference on Theoretical and Mathematical Foundations of Computer Science (TMFCS-10)

We invite draft paper submissions. Please see the website http://www.promoteresearch.org for more details.

Sincerely

John Edward

Publicity committee

RE: Wekalist Digest, Vol 81, Issue 12

2009-11-08T08:05:36Z

Hello,

Does anyone know whether Weka support LSA or not?

If it does, how can I use it?

Best regards,

> From: wekalist-request@...
> Subject: Wekalist Digest, Vol 81, Issue 12
> To: wekalist@...
> Date: Sat, 7 Nov 2009 21:14:06 -0800
>
> 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. Re: Specifying Folds for Leave One Out Cross Validation
> (Peter Reutemann)
> 2. Re: Changing Attribute for Classifying (E_n_r_i_c_O)
> 3. RE: Changing Attribute for Classifying (enrico kusnady)
> 4. Re: Specifying Folds for Leave One Out Cross Validation
> (susan mckeever)
> 5. Are we supposed to Mention CLASS in TESTING DATA (All Good)
> 6. Re: Neural Network implementation in Weka (Feras)
> 7. Re: Are we supposed to Mention CLASS in TESTING DATA
> (Peter Reutemann)
>
>
> ----------------------------------------------------------------------
>
> Message: 1
> Date: Sat, 7 Nov 2009 10:22:44 +1300
> From: Peter Reutemann <fracpete@...>
> Subject: Re: [Wekalist] Specifying Folds for Leave One Out Cross
> Validation
> To: "Weka machine learning workbench list."
> <wekalist@...>
> Message-ID:
> <548e07050911061322o3f45a0b4ib73b614c42cbb9bd@...>
> Content-Type: text/plain; charset=ISO-8859-1
>
> > I want to use Weka on a dataset that consists of 26 days of timestamped
> > activity data from a smart home. I'd like to run "Leave one out" cross
> > validation, where one is a single day's worth of data. I would then like
> > to follow that experiment with with 'Leave two days out' cross validation
> > and so on. Does Weka have some way of letting me specify the contents of
> > cross validation folds to support what I want to do? Or will I have to
> > manually use Training and Test file options under the Explorer Classify to
> > run each fold of the tests manually?
>
> I reckon, you will have to use your own evaluation, as Weka doesn't
> allow you to pick a certain values for evaluation. The groovy script
> "CustomCV.groovy" in the "Downloads" section of the following wiki
> article might be helpful:
> http://weka.wikispaces.com/Using+Weka+from+Groovy
>
> Cheers, Peter
> --
> Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
> http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174
>
>
>
> ------------------------------
>
> Message: 2
> Date: Fri, 6 Nov 2009 14:08:55 -0800 (PST)
> From: E_n_r_i_c_O <rico_104@...>
> Subject: Re: [Wekalist] Changing Attribute for Classifying
> To: wekalist@...
> Message-ID: <26230909.post@...>
> Content-Type: text/plain; charset=us-ascii
>
>
> Hello Peter,
>
> Thank you for your fast answer.
>
> Now it worked just like i wanted.
>
>
>
> bye,
>
>
> Enrico
>
>
>
> --
> View this message in context: http://old.nabble.com/Changing-Attribute-for-Classifying-tp26230856p26230909.html
> Sent from the WEKA mailing list archive at Nabble.com.
>
>
>
>
> ------------------------------
>
> Message: 3
> Date: Fri, 6 Nov 2009 22:08:14 +0000
> From: enrico kusnady <rico_104@...>
> Subject: RE: [Wekalist] Changing Attribute for Classifying
> To: <wekalist@...>
> Message-ID: <BAY114-W1875E66F81F21125A1741AB2AF0@...>
> Content-Type: text/plain; charset="iso-8859-1"
>
>
> Hello Peter,
> Thank you for your fast answer.
> Now it worked just like i wanted.
>
>
> bye,
>
> Enrico
>
>
>
>
>
>
>
>
>
>
> > From: fracpete@...
> > Date: Sat, 7 Nov 2009 10:13:18 +1300
> > Subject: Re: [Wekalist] Changing Attribute for Classifying
> > To: wekalist@...
> >
> > > i need another help from you again.
> > > In the classify i want to change Attribut from (Num) rain_sum as we could
> > > see in this pic
> > > http://old.nabble.com/file/p26230856/1.jpg
> > >
> > > into (Nom) device5_wh like this
> > > http://old.nabble.com/file/p26230856/2.jpg
> > >
> > > Can somebody please tell me how i can change the attribute??
> >
> > I'm not quite sure what your problem is (if it's not about selecting a
> > different attribute from a combobox)...
> >
> > Are you trying to change the class attribute from your own code? If
> > so, use the "setClassIndex(int)" method of the weka.core.Instances
> > class.
> >
> > Cheers, Peter
> > --
> > Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
> > http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174
> >
> > _______________________________________________
> > Wekalist mailing list
> > Send posts to: Wekalist@...
> > 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
>
> _________________________________________________________________
> Windows Live: Friends get your Flickr, Yelp, and Digg updates when they e-mail you.
> http://www.microsoft.com/windows/windowslive/see-it-in-action/social-network-basics.aspx?ocid=PID23461::T:WLMTAGL:ON:WL:en-id:SI_SB_3:092010
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: https://list.scms.waikato.ac.nz/pipermail/wekalist/attachments/20091106/9c898f63/attachment-0001.html
>
> ------------------------------
>
> Message: 4
> Date: Sat, 7 Nov 2009 09:21:14 +0000
> From: susan mckeever <suemckeever@...>
> Subject: Re: [Wekalist] Specifying Folds for Leave One Out Cross
> Validation
> To: "Weka machine learning workbench list."
> <wekalist@...>
> Message-ID:
> <6646ae820911070121r4d4f82f2ib576fad3b9e8a005@...>
> Content-Type: text/plain; charset="iso-8859-1"
>
> Thanks Peter
>
>
>
> On Fri, Nov 6, 2009 at 9:22 PM, Peter Reutemann <fracpete@...> wrote:
>
> > > I want to use Weka on a dataset that consists of 26 days of timestamped
> > > activity data from a smart home. I'd like to run "Leave one out" cross
> > > validation, where one is a single day's worth of data. I would then
> > like
> > > to follow that experiment with with 'Leave two days out' cross
> > validation
> > > and so on. Does Weka have some way of letting me specify the contents of
> > > cross validation folds to support what I want to do? Or will I have to
> > > manually use Training and Test file options under the Explorer Classify
> > to
> > > run each fold of the tests manually?
> >
> > I reckon, you will have to use your own evaluation, as Weka doesn't
> > allow you to pick a certain values for evaluation. The groovy script
> > "CustomCV.groovy" in the "Downloads" section of the following wiki
> > article might be helpful:
> > http://weka.wikispaces.com/Using+Weka+from+Groovy
> >
> > Cheers, Peter
> > --
> > Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
> > http://www.cs.waikato.ac.nz/~fracpete/<http://www.cs.waikato.ac.nz/%7Efracpete/> Ph. +64 (7) 858-5174
> >
> > _______________________________________________
> > Wekalist mailing list
> > Send posts to: Wekalist@...
> > 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<http://www.cs.waikato.ac.nz/%7Eml/weka/mailinglist_etiquette.html>
> >
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: https://list.scms.waikato.ac.nz/pipermail/wekalist/attachments/20091107/7fd469c3/attachment-0001.html
>
> ------------------------------
>
> Message: 5
> Date: Sat, 7 Nov 2009 13:52:07 +0100
> From: All Good <kimskams80@...>
> Subject: [Wekalist] Are we supposed to Mention CLASS in TESTING DATA
> To: Weka List <wekalist@...>
> Message-ID:
> <114cd5e90911070452icc6db40g2df46f1a62fb20dd@...>
> Content-Type: text/plain; charset="iso-8859-1"
>
> Sorry for this almost stupid question .. But its just for confirmation that
> in the testing file, are we suppoed to mention the class label after each
> instance??
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: https://list.scms.waikato.ac.nz/pipermail/wekalist/attachments/20091107/63ed0ff3/attachment-0001.html
>
> ------------------------------
>
> Message: 6
> Date: Sat, 7 Nov 2009 05:10:14 -0800 (PST)
> From: Feras <feras_o@...>
> Subject: Re: [Wekalist] Neural Network implementation in Weka
> To: "Weka machine learning workbench list."
> <wekalist@...>
> Message-ID: <777929.14475.qm@...>
> Content-Type: text/plain; charset="iso-8859-1"
>
>
> Thanks Peter, you are doing a great job
>
> Feras
>
> --- On Fri, 11/6/09, Peter Reutemann <fracpete@...> wrote:
>
>
> From: Peter Reutemann <fracpete@...>
> Subject: Re: [Wekalist] Neural Network implementation in Weka
> To: "Weka machine learning workbench list." <wekalist@...>
> Date: Friday, November 6, 2009, 5:10 PM
>
>
> > what is the name of the algorithm to use Neural Network in Weka?
>
> weka.classifiers.functions.MultilayerPerceptron
>
> Cheers, Peter
> --
> Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
> http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174
>
> _______________________________________________
> Wekalist mailing list
> Send posts to: Wekalist@...
> 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
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: https://list.scms.waikato.ac.nz/pipermail/wekalist/attachments/20091107/d2171f8a/attachment-0001.html
>
> ------------------------------
>
> Message: 7
> Date: Sun, 8 Nov 2009 18:12:10 +1300
> From: Peter Reutemann <fracpete@...>
> Subject: Re: [Wekalist] Are we supposed to Mention CLASS in TESTING
> DATA
> To: "Weka machine learning workbench list."
> <wekalist@...>
> Message-ID:
> <548e07050911072112j6b5c8760ue1fa7ed5cf2c5734@...>
> Content-Type: text/plain; charset=ISO-8859-1
>
> > Sorry for this almost stupid question .. But its just for confirmation that
> > in the testing file, are we suppoed to mention the class label after each
> > instance??
>
> If your test set doesn't contain any values for the class attribute,
> then you can't evaluate the performance of a classifier (classifiers
> aren't psychic yet). Even if you just want to output the predictions,
> your test set still needs the same structure as the training set. In
> that case, just use missing values (in ARFF files, missing values are
> denoted by the question mark).
>
> Cheers, Peter
> --
> Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
> http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174
>
>
>
> ------------------------------
>
> _______________________________________________
> Wekalist mailing list
> Wekalist@...
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
>
>
> End of Wekalist Digest, Vol 81, Issue 12
> ****************************************

Windows Live: Make it easier for your friends to see what you’re up to on Facebook.
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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

Weka ensemble

2009-11-08T03:23:41Z

dear all,

I am working on an ensemble algorithm to combine classifiers to improve
the classification accuracy. I found that many Weka's classes such as
Boosting, Bagging and etc. which seem quite relevant to me. Besides that,
I think I've seen people around writing java code to make use of these
classes. Does anybody knows or can point me locations that can run the
algorithms from java code? Is the same way as in
http://weka.wikispaces.com/Programmatic+Use can be used to run weka's
ensemble algorithms?

kind regards
azizia.

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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: Are we supposed to Mention CLASS in TESTING DATA

2009-11-08T02:34:46Z

Please no top-posting, see mailing list etiquette why
(http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html).

> Writing a text analysis program and generating test datasets with
> StringToWordVector to classify with a pre-trained classifier, I noticed that
> it classified even without the exact same format.

Use batch filtering (see FAQ "How do I generate compatible train and
test sets that get processed with a filter?") or the
FilteredClassifier meta-classifier (StringToWordVector and your choice
of base-classifier) to ensure compatible datasets.

> I am now under the impression that Weka will match up the columns with
> similar names, and consider the ones that don't exist are ignored. Maybe
> they're counted as 'missing' by the classifier? I don't really know! I just
> know that it works!

Weka *assumes* that the datasets have the same structure. There is no
magic happening, Weka's classifiers will just use attribute indices
when making predictions. If they datasets differ, values from
different attributes (at the same location) will be picked. In a
nutshell: the results cannot be trusted.

> Is it just taking the first n attributes, and discarding n+1 and on?

That depends on the classifier. A decision tree based classifier might
only need 5 attributes out of a 1000.

> Incorrectly comparing attributes that aren't the same metric?

Weka's internal data format are just doubles (the indices of labels
are stored for nominal attributes), nothing more. Simple and fast
number comparisons happen internally. Computational expensive mappings
original attribute space of input and potentially different attribute
space when predicting would make it unbearably slow.

> (if so, is
> there a way to StringToWordVector a String field into the appropriate
> attribute set?)

Like I said above, use batch filtering or the FilteredClassifier approach.

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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: Are we supposed to Mention CLASS in TESTING DATA

2009-11-08T02:23:27Z

Writing a text analysis program and generating test datasets with StringToWordVector to classify with a pre-trained classifier, I noticed that it classified even without the exact same format.

I am now under the impression that Weka will match up the columns with similar names, and consider the ones that don't exist are ignored. Maybe they're counted as 'missing' by the classifier? I don't really know! I just know that it works!

Is it just taking the first n attributes, and discarding n+1 and on? Incorrectly comparing attributes that aren't the same metric? (if so, is there a way to StringToWordVector a String field into the appropriate attribute set?)

On Sun, Nov 8, 2009 at 12:12 AM, Peter Reutemann <fracpete@...> wrote:

> Sorry for this almost stupid question .. But its just for confirmation that
> in the testing file, are we suppoed to mention the class label after each
> instance??

If your test set doesn't contain any values for the class attribute,
then you can't evaluate the performance of a classifier (classifiers
aren't psychic yet). Even if you just want to output the predictions,
your test set still needs the same structure as the training set. In
that case, just use missing values (in ARFF files, missing values are
denoted by the question mark).

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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 Linear Regression Problem

2009-11-08T00:08:47Z

Hi,

We have a set of 25 Topics. We use 10 Topics for training and 10 for Testing and we kept 5 for confirmation of the results with 3 DIFERENT Combinations of FEATURES (so that we can show what features are more important).

The results are same as we expecting except one strange exception is occuring. As we expecting, the result of Comb2 should be larger than Comb1 which is demonstrated when testing on 5 Topics (for confirmation) but its less on TESTING DATA (ie Comb2 < Comb1) .

Do you think its problem of Less Training Data or what??

thanks in adv

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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: some variables dropped out, could not be fully loaded

2009-11-07T23:16:07Z

> I use weka in windows, and I got a set of data in linux, and
> save the data file in csv file format, the data set file has 342
> columes, that is, the data set have 342 variables , the samples have
> 75 instances,
> but in windows, the excel could not load all variables, it just
> load 256 columes, and in windows, weka also just load 256 varialbes,
> drop out the rest,
>
>
> how to find out the dropped variables in windows, what should I set in
> weka

Weka does *not* have a limit of columns (what version of Weka are you
using, anyway?). Something else must have gone wrong there. Maybe you
accidentally saved the CSV file with Excel and that deleted all the
other columns?

BTW OpenOffice 3 handles up to 1024 columns:
http://en.wikipedia.org/wiki/OpenOffice.org_Calc#Specifications

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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

some variables dropped out, could not be fully loaded

2009-11-07T22:56:35Z

Hi
I use weka in windows, and I got a set of data in linux, and
save the data file in csv file format, the data set file has 342
columes, that is, the data set have 342 variables , the samples have
75 instances,
but in windows, the excel could not load all variables, it just
load 256 columes, and in windows, weka also just load 256 varialbes,
drop out the rest,

how to find out the dropped variables in windows, what should I set in
weka
Could you help me
Thank you

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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: Are we supposed to Mention CLASS in TESTING DATA

2009-11-07T21:12:10Z

> Sorry for this almost stupid question .. But its just for confirmation that
> in the testing file, are we suppoed to mention the class label after each
> instance??

If your test set doesn't contain any values for the class attribute,
then you can't evaluate the performance of a classifier (classifiers
aren't psychic yet). Even if you just want to output the predictions,
your test set still needs the same structure as the training set. In
that case, just use missing values (in ARFF files, missing values are
denoted by the question mark).

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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: Neural Network implementation in Weka

2009-11-07T05:10:14Z

Thanks Peter, you are doing a great job

Feras

--- On Fri, 11/6/09, Peter Reutemann <fracpete@...> wrote:

From: Peter Reutemann <fracpete@...>
Subject: Re: [Wekalist] Neural Network implementation in Weka
To: "Weka machine learning workbench list." <wekalist@...>
Date: Friday, November 6, 2009, 5:10 PM

> what is the name of the algorithm to use Neural Network in Weka?

weka.classifiers.functions.MultilayerPerceptron

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
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