Indra M | 2 Sep 08:02 2015
Picon

(no subject)

Hi

Greetings.

I would like to know how to use WEKA for regression problem.

 I have attached two files : one is training data and To_be_predicted_data.

The variables are
water_vol, temp, humidity,
concentration,
height, strength.



In the above variables, strength is target variable and rest are independent variables.

I would like to know How to use multiple regression model in WEKA for the sample data? Please tell the steps for the same.

Also how can we use the ML algorithms for this regression problem.

Kindly advice on this.

Thanks  in advance.



Regards
Indrajit
Attachment (To_be_predicted_data.xlsx): application/vnd.openxmlformats-officedocument.spreadsheetml.sheet, 11 KiB
Attachment (traininng data.xlsx): application/vnd.openxmlformats-officedocument.spreadsheetml.sheet, 12 KiB
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Eibe Frank | 1 Sep 22:38 2015
Picon
Picon

Re: filters

No, there is a separate filter called SMOTE.

Cheers,
Eibe

> On 2/09/2015, at 2:17 am, Randa Majalli <randao <at> hu.edu.jo> wrote:
> 
> Dear Eibe,
> 
> I wanted to know if the resample filter uses SMOTE? 
> 
> Thanks a lot,
> Randa, 
> 
> On Mon, Aug 31, 2015 at 2:25 AM, Eibe Frank <eibe <at> waikato.ac.nz> wrote:
> Here is the reference for SMOTE, from the Javadoc (also available under “Global info” in the GUI):
> 
> BibTeX:
> * <pre>
> * &#64;article{al.2002,
> *    author = {Nitesh V. Chawla et. al.},
> *    journal = {Journal of Artificial Intelligence Research},
> *    pages = {321-357},
> *    title = {Synthetic Minority Over-sampling Technique},
> *    volume = {16},
> *    year = {2002}
> * }
> * </pre>
> * <p/>
> 
> For the other filters, there is no better description than what’s in the online help. A similarly
high-level description is in our data mining book.
> 
> Cheers,
> Eibe
> 
> > On 20/08/2015, at 11:17 pm, Randa Majalli <randao <at> hu.edu.jo> wrote:
> >
> > I hope my message finds you well,
> >
> > Actually I am working on an imbalanced data set with the following distribution of instances:
> >
> > Class
> > -ve
> > +ve
> > instances
> > 13725
> > 3090
> >
> >
> > ​I used different resampling techniques in weka:
> >
> > 1. filters.supervised.instance.smote​
> > ​2. filters.supervised.instance.spreadsubsample​
> > 3. Mix of oversampling and undersampling: using the following filter in weka:
filter.supervised.instance.resample, with the following parameters specified for resample:
> >
> > o   Bias to uniform class: 1.0
> >
> > o   Invert selection: false
> >
> > o   No replacement: false
> >
> > o   Random seed: 1.0
> >
> > o   Random size percent: 100
> >
> > ​Actually, my question is the following: no references were provided that explain how does resample
filter works? could you please provide me with such references ASAP,
> >
> > Best Regards,​
> >
> > ​Randa,​
> >
> > _______________________________________________
> > Wekalist mailing list
> > Send posts to: Wekalist <at> list.waikato.ac.nz
> > List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
> > List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> 
> _______________________________________________
> Wekalist mailing list
> Send posts to: Wekalist <at> list.waikato.ac.nz
> List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> 
> 
> 
> -- 
> Randa O. Mujalli, PhD
> 
> Assistant Professor
> 
> The Hashemite University
> 
> Department of Civil Engineering 
> 
> P.O. Box 150459
> 13115-Zarqa 
> 
> Telf: +962 5 390 3333 - 4777
> 
> Fax: +962 5 382 6348
> e-mail: randao <at> hu.edu.jo 
> 
> _______________________________________________
> Wekalist mailing list
> Send posts to: Wekalist <at> list.waikato.ac.nz
> List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Indra M | 1 Sep 12:33 2015
Picon

Predicting the unknown class

Hi everyone,

I have a query to ask to the Weka community.


Lets say I have a training file (say F1) with 10 attributes (say a1, a2, a3,.. ..  ...a9, a10) and a binary class C1(say 0 and 1).


Lets say I have training file (F1) with 1000 samples.


I have lets say a test file (F2) consisting of unknown class with 20 samples where the class is unknown. We need to determine the class for this 20 samples. This file contains all the 10 variables (as in F1) and class column is empty.



Can you suggest me how to set the configuration in weka? Please tell the steps clearly.

I have attached the training and the test file for reference.


Thanks in advance.

--
Indrajit



Attachment (My_Test_File_.arff): application/octet-stream, 1090 bytes
Attachment (Train_File.arff): application/octet-stream, 6439 bytes
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Razer Anthom | 1 Sep 05:00 2015
Picon

Problems with GridSearch 1.0.8 and Weka 3.7.12


   Hi all...

   I am trying to run GridSearch with SMOReg in command-line, but it is not working. The command is:

java -classpath ./lib/weka.jar:./lib/gridSearch.jar weka.classifiers.meta.GridSearch -E CC -y-property classifier.gamma -y-min -5.0 -y-max 2.0 -y-step 1.0 -y-base 10.0 -y-expression "pow(BASE,I)" -x-property classifier.cost -x-min 1.0 -x-max 20.0 -x-step 1.0 -x-base 10.0 -x-expression "I" -sample-size 100.0 -traversal COLUMN-WISE -log-file "./" -S 1 -t train.arff -T test.arff -d grid.model -W  weka.classifiers.functions.SMOreg -- -C 1.0 -N 0 -I "weka.classifiers.functions.supportVector.RegSMOImproved -T 0.001 -V -P 1.0E-12 -L 0.001 -W 1" -K "weka.classifiers.functions.supportVector.RBFKernel -G 0.1 -C 250007"


    Libs are: WEKA 3.7.12 and GRIDSEARCH 1.0.8 (compatible, by documentation) and they are in ./lib folder. Error obtained is:


java.lang.NoClassDefFoundError: weka/core/MathematicalExpression
at weka.classifiers.meta.GridSearch$SetupGenerator.evaluate(GridSearch.java:1748)
at weka.classifiers.meta.GridSearch$EvaluationTask.run(GridSearch.java:1900)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615)
at java.lang.Thread.run(Thread.java:745)

   Inside weka.jar there is not this class. I tried others combinations of versions, like GridSearch 1.0.7, but then it complains of PLSFilter, even if I put "-filter weka.filters.AllFilter" parameter.

   Do you have any suggestions?

   Thank you in advance

Razer.

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
psgheptium | 1 Sep 00:53 2015

Which would you recommend Weka 3.6 or 3.7? - development for online ML system

Hi, 

would you recommend using Weka 3.7.x in a production code? 

It is stated that it has the same core as 3.6.x but also extra experimental
features (such as DenseInstance). 

Is there any differences I should really consider before using dev version?

Would very much appreciate your answers. Sorry for not so much technical
question, hope it's not a matter of opinion, but informed choice. :)

Kind regards!

--
View this message in context: http://weka.8497.n7.nabble.com/Which-would-you-recommend-Weka-3-6-or-3-7-development-for-online-ML-system-tp35412.html
Sent from the WEKA mailing list archive at Nabble.com.
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html

Álvaro Gómez Uría | 31 Aug 15:13 2015
Picon

Wrapper in WEKA

Hello,
i am wondering if is there any website where I can read about the technique Wrapper in WEKA. How is implemented (pseudocode) or something similar?.

In case the answer is no, I have a question: for choosing the correct subset of features, WEKA divided the data in training and test as well?. In that case, how is possible that in 1 fold maybe one features is representative but no in other fold? is the average?

Thanks,
Alvaro  
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Martin | 29 Aug 04:30 2015
Picon

Re: supervised discretize filter is not active

I’m not sure whether the conversion process that you performed is correct or not (conversion tools), but I still believe that your data attributes are still not consistent (I haven’t tried SAS myself). Try to use another tool to combine all your attributes, like SPSS. From that tool explore your attributes type, and make sure that they are all compatible. To be more precise, explore your data type in SPSS using the option “Variable View” (at the bottom of the SPSS). This option will let you know the type of each column of your data. Make sure that your data have consistent attribute type (you can use the previous attribute type that was only accepted to be load into WEKA previously to give you a clue about major accepted data type). After that, save your SPSS file as CSV directly, or save it as an Excel and then convert it to CSV. This is what I can think of currently without having the data.


On the other hand, the CPU data I provided in my last post is only to help you to know how the data structure can be within the CSV file, if you need to let Apriori handle it, you have to select “No class” from the Preprocess panel (just above the histogram) before you invoke either Discretize filter or NumericToNominal filter, I tried that and it worked fine with both. Note that, after applying the filters you need to make sure that your dataset from the nominal type.


Regards,

Martin


View this message in context: Re: supervised discretize filter is not active
Sent from the WEKA mailing list archive at Nabble.com.
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Martin | 28 Aug 13:09 2015
Picon

Re: supervised discretize filter is not active

Again, looking to your case, I believe the reason behind this can be due to your the class type which I assume it is form the numeric type. Thus, the supervised Discretize cannot handle such type. My suggestion is: either to use "weka.filters.unsupervised.attribute.Discretize" or "weka.filters.unsupervised.attribute.NumericToNominal".

NB: For your information, Association rules mining is unsupervised learning, so if you need to consider the last attribute of your data (class), then you need to select the option "No class". 

Regards, 
Martin

On 28 August 2015 at 18:01, shalom [via WEKA] <[hidden email]> wrote:
I do have .csv file and am using weka 3.6 for Association rule. After I imported the file , to change the numerical to nominal before applying Apriori but supervised attribute filter is disabled.Is the problem is within the data ?

If you reply to this email, your message will be added to the discussion below:
http://weka.8497.n7.nabble.com/supervised-discretize-filter-is-not-active-tp35390.html
To unsubscribe from WEKA, click here.
NAML


View this message in context: Re: supervised discretize filter is not active
Sent from the WEKA mailing list archive at Nabble.com.
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Martin | 28 Aug 12:45 2015
Picon

Re: supervised discretize filter is not active

You may need to provide further information about your dataset type (attribute and class) or sample of it. Make sure that your data are well designed. 

Regards, 
Martin

On 28 August 2015 at 18:01, shalom [via WEKA] <[hidden email]> wrote:
I do have .csv file and am using weka 3.6 for Association rule. After I imported the file , to change the numerical to nominal before applying Apriori but supervised attribute filter is disabled.Is the problem is within the data ?

If you reply to this email, your message will be added to the discussion below:
http://weka.8497.n7.nabble.com/supervised-discretize-filter-is-not-active-tp35389.html
To unsubscribe from WEKA, click here.
NAML


View this message in context: Re: supervised discretize filter is not active
Sent from the WEKA mailing list archive at Nabble.com.
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
alwin | 28 Aug 10:20 2015
Picon

Visualize Tree in WEKA Experimenter

Hi,

I'm using the Paired T-Tester of the WEKA Experimenter to compare the
performance of two models constructed using the J48 classifier. I want to
visualize the final trees derived from the cross validations so that I can
inspect the model.

I know this can be done in the Explorer. However, the results in the WEKA
Explorer slightly deviate from the results in the Experimenter, meaning that
the trees also deviate. In addition, when using the WEKA Explorer, I have to
do the T-test manually.

Is it also possible to viasualize the decision tree in the WEKA
Experimenter? Or do you have any other suggestions how to visualize the
tree?

--
View this message in context: http://weka.8497.n7.nabble.com/Visualize-Tree-in-WEKA-Experimenter-tp35388.html
Sent from the WEKA mailing list archive at Nabble.com.
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html

tien dh | 28 Aug 09:55 2015
Picon

Some questions related to multilayer perceptron parameters

Hello everyone,

I'm doing some experiments with multilayer perceptron and I have some questions. I read check help file but it seems not clear enough for me:

  • What is nominalToBinaryFilter? How to use?

  • normalizeAttribute: I think this is to scale value of features to [-1, 1] range. But how to do it in case the value is not numeric, for example with weather dataset (features have value such as rainy, sunny, etc.)

  • reset: This will reset if the current training process diverges and start again with a lower learning rate. How much should we decrease the current learning rate? (how to identify the next learning rate)

  • Initial weights: This isn't a parameter, but how to initialize initial weights? Is it symmetric (something like values inside [-ε, +ε])?
  • When the training process stop? I think it should stop when either epoch limit reach or cost function (J) small enough, but in the help file, weka consults validation set. What's the role of validation set here?
Thank you.

--
Thanks and Best Regards,

Do Huu Tien (Mr.)
Handphone No: 0493110507
Email: tiendh11986 <at> gmail.com
Y!M: beuforever
Skype: do.huu.tien
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist <at> list.waikato.ac.nz
List info and subscription status: http://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html

Gmane