Measuring accuracy and efficiency of association rules!

classic Classic list List threaded Threaded
49 messages Options
123
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Measuring accuracy and efficiency of association rules!

Bhupesh Rawat
>
> Dear Sir/Madam,
>
   How to measure the Accuracy and Efficiency(running time and space)
of association
   rules discovered through WEKA.




--
Thanks & Regards
Bhupesh Rawat.
Ph.D Scholar
Department of Computer Science,Babasaheb Bhimrao Ambedkar University
Vidya Vihar,Rai Bareilly road(Lucknow)
Ph. No: +91-9897065948

...........................................................................................................................
*A man is the best judge of himself and he has to pay the price for what he
does.*
...........................................................................................................................
_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Eibe Frank-2
Administrator
The running time is output under “Elapsed time” when you run Apriori.

The size of the final Apriori model as a serialised Java object can be established saving it to a file and considering the file size. Note that this is different from the size of the object in memory (see, e.g., http://stackoverflow.com/questions/7146559/serialized-object-size-vs-in-memory-object-size-in-java#7146941).

I don’t know of a good way to measure peak memory consumption of a Java program (after garbage collection). A crude way would be to run the program from the command-line (to avoid overhead associated with the GUIs) with different maximum heap sizes, e.g., increasing the heap size until the program runs through. Another option is to look at the heap size in a profiler (e.g., visualvm), enforcing garbage collection before a readout.

Cheers,
Eibe

> On 11/02/2017, at 10:27 PM, Bhupesh Rawat <[hidden email]> wrote:
>
>>
>> Dear Sir/Madam,
>>
>   How to measure the Accuracy and Efficiency(running time and space)
> of association
>   rules discovered through WEKA.
>
>
>
>
> --
> Thanks & Regards
> Bhupesh Rawat.
> Ph.D Scholar
> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
> Vidya Vihar,Rai Bareilly road(Lucknow)
> Ph. No: +91-9897065948
>
> ...........................................................................................................................
> *A man is the best judge of himself and he has to pay the price for what he
> does.*
> ...........................................................................................................................
> _______________________________________________
> Wekalist mailing list
> Send posts to: [hidden email]
> List info and subscription status: https://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: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Peter Reutemann
> The size of the final Apriori model as a serialised Java object can be established saving it to a file and considering the file size. Note that this is different from the size of the object in memory (see, e.g., http://stackoverflow.com/questions/7146559/serialized-object-size-vs-in-memory-object-size-in-java#7146941).
>
> I don’t know of a good way to measure peak memory consumption of a Java program (after garbage collection). A crude way would be to run the program from the command-line (to avoid overhead associated with the GUIs) with different maximum heap sizes, e.g., increasing the heap size until the program runs through. Another option is to look at the heap size in a profiler (e.g., visualvm), enforcing garbage collection before a readout.

You can use the sizeofag javaagent for determining the size of a Java object:
https://github.com/fracpete/sizeofag

Credits to Maxim Zakharenkov, who wrote the original code.

Cheers, Peter
--
Peter Reutemann
Dept. of Computer Science
University of Waikato, NZ
+64 (7) 858-5174
http://www.cms.waikato.ac.nz/~fracpete/
http://www.data-mining.co.nz/
_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Bhupesh Rawat
Dear Sir/Madam

I have discovered some rules through weka. Could you tell me how to measure  the accuracy of those rules.







On Tue, Feb 14, 2017 at 3:44 AM, Peter Reutemann <[hidden email]> wrote:
> The size of the final Apriori model as a serialised Java object can be established saving it to a file and considering the file size. Note that this is different from the size of the object in memory (see, e.g., http://stackoverflow.com/questions/7146559/serialized-object-size-vs-in-memory-object-size-in-java#7146941).
>
> I don’t know of a good way to measure peak memory consumption of a Java program (after garbage collection). A crude way would be to run the program from the command-line (to avoid overhead associated with the GUIs) with different maximum heap sizes, e.g., increasing the heap size until the program runs through. Another option is to look at the heap size in a profiler (e.g., visualvm), enforcing garbage collection before a readout.

You can use the sizeofag javaagent for determining the size of a Java object:
https://github.com/fracpete/sizeofag

Credits to Maxim Zakharenkov, who wrote the original code.

Cheers, Peter
--
Peter Reutemann
Dept. of Computer Science
University of Waikato, NZ
+64 (7) 858-5174
http://www.cms.waikato.ac.nz/~fracpete/
http://www.data-mining.co.nz/
_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html



--
Thanks & Regards
Bhupesh Rawat.
Ph.D Scholar
Department of Computer Science,Babasaheb Bhimrao Ambedkar University
Vidya Vihar,Rai Bareilly road(Lucknow)
Ph. No: +91-9897065948

...........................................................................................................................
*A man is the best judge of himself and he has to pay the price for what he
does.*
...........................................................................................................................




_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Eibe Frank-2
Administrator
You mean beyond confidence, lift, or one the other metrics that you can get in the output of each rule? This is a tough question. One way may be to use the association rule mining algorithm to build classification rules and then evaluate the accuracy of those classification rules. We had a paper on this quite a while back:

Mutter, S., Hall, M., & Frank, E. (2004, December). Using classification to evaluate the output of confidence-based association rule mining. In Australasian Joint Conference on Artificial Intelligence (pp. 538-549). Springer Berlin Heidelberg.

I suppose you could also evaluate the individual association rules on a separate test set, by computing the confidence measure, etc., on the test set for each rule, but this functionality is not provided by WEKA.

Cheers,
Eibe

> On 23/02/2017, at 12:46 AM, Bhupesh Rawat <[hidden email]> wrote:
>
> Dear Sir/Madam
>
> I have discovered some rules through weka. Could you tell me how to measure  the accuracy of those rules.
>
>
>
>
>
>
>
> On Tue, Feb 14, 2017 at 3:44 AM, Peter Reutemann <[hidden email]> wrote:
> > The size of the final Apriori model as a serialised Java object can be established saving it to a file and considering the file size. Note that this is different from the size of the object in memory (see, e.g., http://stackoverflow.com/questions/7146559/serialized-object-size-vs-in-memory-object-size-in-java#7146941).
> >
> > I don’t know of a good way to measure peak memory consumption of a Java program (after garbage collection). A crude way would be to run the program from the command-line (to avoid overhead associated with the GUIs) with different maximum heap sizes, e.g., increasing the heap size until the program runs through. Another option is to look at the heap size in a profiler (e.g., visualvm), enforcing garbage collection before a readout.
>
> You can use the sizeofag javaagent for determining the size of a Java object:
> https://github.com/fracpete/sizeofag
>
> Credits to Maxim Zakharenkov, who wrote the original code.
>
> Cheers, Peter
> --
> Peter Reutemann
> Dept. of Computer Science
> University of Waikato, NZ
> +64 (7) 858-5174
> http://www.cms.waikato.ac.nz/~fracpete/
> http://www.data-mining.co.nz/
> _______________________________________________
> Wekalist mailing list
> Send posts to: [hidden email]
> List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>
>
>
> --
> Thanks & Regards
> Bhupesh Rawat.
> Ph.D Scholar
> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
> Vidya Vihar,Rai Bareilly road(Lucknow)
> Ph. No: +91-9897065948
>
> ...........................................................................................................................
> *A man is the best judge of himself and he has to pay the price for what he
> does.*
> ...........................................................................................................................
>
>
>
> _______________________________________________
> Wekalist mailing list
> Send posts to: [hidden email]
> List info and subscription status: https://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: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Bhupesh Rawat

Thank you so much for the response!!

On Feb 23, 2017 8:26 AM, "Eibe Frank" <[hidden email]> wrote:
You mean beyond confidence, lift, or one the other metrics that you can get in the output of each rule? This is a tough question. One way may be to use the association rule mining algorithm to build classification rules and then evaluate the accuracy of those classification rules. We had a paper on this quite a while back:

Mutter, S., Hall, M., & Frank, E. (2004, December). Using classification to evaluate the output of confidence-based association rule mining. In Australasian Joint Conference on Artificial Intelligence (pp. 538-549). Springer Berlin Heidelberg.

I suppose you could also evaluate the individual association rules on a separate test set, by computing the confidence measure, etc., on the test set for each rule, but this functionality is not provided by WEKA.

Cheers,
Eibe

> On 23/02/2017, at 12:46 AM, Bhupesh Rawat <[hidden email]> wrote:
>
> Dear Sir/Madam
>
> I have discovered some rules through weka. Could you tell me how to measure  the accuracy of those rules.
>
>
>
>
>
>
>
> On Tue, Feb 14, 2017 at 3:44 AM, Peter Reutemann <[hidden email]> wrote:
> > The size of the final Apriori model as a serialised Java object can be established saving it to a file and considering the file size. Note that this is different from the size of the object in memory (see, e.g., http://stackoverflow.com/questions/7146559/serialized-object-size-vs-in-memory-object-size-in-java#7146941).
> >
> > I don’t know of a good way to measure peak memory consumption of a Java program (after garbage collection). A crude way would be to run the program from the command-line (to avoid overhead associated with the GUIs) with different maximum heap sizes, e.g., increasing the heap size until the program runs through. Another option is to look at the heap size in a profiler (e.g., visualvm), enforcing garbage collection before a readout.
>
> You can use the sizeofag javaagent for determining the size of a Java object:
> https://github.com/fracpete/sizeofag
>
> Credits to Maxim Zakharenkov, who wrote the original code.
>
> Cheers, Peter
> --
> Peter Reutemann
> Dept. of Computer Science
> University of Waikato, NZ
> +64 (7) 858-5174
> http://www.cms.waikato.ac.nz/~fracpete/
> http://www.data-mining.co.nz/
> _______________________________________________
> Wekalist mailing list
> Send posts to: [hidden email]
> List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>
>
>
> --
> Thanks & Regards
> Bhupesh Rawat.
> Ph.D Scholar
> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
> Vidya Vihar,Rai Bareilly road(Lucknow)
> Ph. No: +91-9897065948
>
> ...........................................................................................................................
> *A man is the best judge of himself and he has to pay the price for what he
> does.*
> ...........................................................................................................................
>
>
>
> _______________________________________________
> Wekalist mailing list
> Send posts to: [hidden email]
> List info and subscription status: https://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: [hidden email]
List info and subscription status: https://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: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Bhupesh Rawat
I have a small dataset which contains student enrolment data in
various courses. If a student has selected a particular course it is
indicated by ‘Y’ else ‘N’ is used. I have also attached a file for
better understanding of the dataset. I am interested in knowing if it
is possible to measure the accuracy of the association rules with this
dataset by the proposed approach in your paper.

On 2/23/17, Bhupesh Rawat <[hidden email]> wrote:

> Thank you so much for the response!!
> On Feb 23, 2017 8:26 AM, "Eibe Frank" <[hidden email]> wrote:
>
>> You mean beyond confidence, lift, or one the other metrics that you can
>> get in the output of each rule? This is a tough question. One way may be
>> to
>> use the association rule mining algorithm to build classification rules
>> and
>> then evaluate the accuracy of those classification rules. We had a paper
>> on
>> this quite a while back:
>>
>> Mutter, S., Hall, M., & Frank, E. (2004, December). Using classification
>> to evaluate the output of confidence-based association rule mining. In
>> Australasian Joint Conference on Artificial Intelligence (pp. 538-549).
>> Springer Berlin Heidelberg.
>>
>> I suppose you could also evaluate the individual association rules on a
>> separate test set, by computing the confidence measure, etc., on the test
>> set for each rule, but this functionality is not provided by WEKA.
>>
>> Cheers,
>> Eibe
>>
>> > On 23/02/2017, at 12:46 AM, Bhupesh Rawat <[hidden email]> wrote:
>> >
>> > Dear Sir/Madam
>> >
>> > I have discovered some rules through weka. Could you tell me how to
>> measure  the accuracy of those rules.
>> >
>> >
>> >
>> >
>> >
>> >
>> >
>> > On Tue, Feb 14, 2017 at 3:44 AM, Peter Reutemann
>> > <[hidden email]>
>> wrote:
>> > > The size of the final Apriori model as a serialised Java object can
>> > > be
>> established saving it to a file and considering the file size. Note that
>> this is different from the size of the object in memory (see, e.g.,
>> http://stackoverflow.com/questions/7146559/serialized-
>> object-size-vs-in-memory-object-size-in-java#7146941).
>> > >
>> > > I don’t know of a good way to measure peak memory consumption of a
>> Java program (after garbage collection). A crude way would be to run the
>> program from the command-line (to avoid overhead associated with the
>> GUIs)
>> with different maximum heap sizes, e.g., increasing the heap size until
>> the
>> program runs through. Another option is to look at the heap size in a
>> profiler (e.g., visualvm), enforcing garbage collection before a readout.
>> >
>> > You can use the sizeofag javaagent for determining the size of a Java
>> object:
>> > https://github.com/fracpete/sizeofag
>> >
>> > Credits to Maxim Zakharenkov, who wrote the original code.
>> >
>> > Cheers, Peter
>> > --
>> > Peter Reutemann
>> > Dept. of Computer Science
>> > University of Waikato, NZ
>> > +64 (7) 858-5174
>> > http://www.cms.waikato.ac.nz/~fracpete/
>> > http://www.data-mining.co.nz/
>> > _______________________________________________
>> > Wekalist mailing list
>> > Send posts to: [hidden email]
>> > List info and subscription status: https://list.waikato.ac.nz/
>> mailman/listinfo/wekalist
>> > List etiquette: http://www.cs.waikato.ac.nz/~
>> ml/weka/mailinglist_etiquette.html
>> >
>> >
>> >
>> > --
>> > Thanks & Regards
>> > Bhupesh Rawat.
>> > Ph.D Scholar
>> > Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>> > Vidya Vihar,Rai Bareilly road(Lucknow)
>> > Ph. No: +91-9897065948
>> >
>> > ............................................................
>> ...............................................................
>> > *A man is the best judge of himself and he has to pay the price for
>> > what
>> he
>> > does.*
>> > ............................................................
>> ...............................................................
>> >
>> >
>> >
>> > _______________________________________________
>> > Wekalist mailing list
>> > Send posts to: [hidden email]
>> > List info and subscription status: https://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: [hidden email]
>> List info and subscription status: https://list.waikato.ac.nz/
>> mailman/listinfo/wekalist
>> List etiquette: http://www.cs.waikato.ac.nz/~
>> ml/weka/mailinglist_etiquette.html
>>
>

--
Thanks & Regards
Bhupesh Rawat.
Ph.D Scholar
Department of Computer Science,Babasaheb Bhimrao Ambedkar University
Vidya Vihar,Rai Bareilly road(Lucknow)
Ph. No: +91-9897065948

...........................................................................................................................
*A man is the best judge of himself and he has to pay the price for what he
does.*
...........................................................................................................................

_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html

students' data after preprocessin.xlsx (16K) Download Attachment
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Eibe Frank-2
Administrator
No, not really. However, the dataset is quite small. You could just run a classification rule learner such as JRip or PART on the data, treating each of the attributes in turn as the class attribute. Then you can estimate classification accuracy using cross-validation.

You could also create combinations of attributes using the CartesionProduct filter.

Cheers,
Eibe

> On 24/02/2017, at 3:11 AM, Bhupesh Rawat <[hidden email]> wrote:
>
> I have a small dataset which contains student enrolment data in
> various courses. If a student has selected a particular course it is
> indicated by ‘Y’ else ‘N’ is used. I have also attached a file for
> better understanding of the dataset. I am interested in knowing if it
> is possible to measure the accuracy of the association rules with this
> dataset by the proposed approach in your paper.
>
> On 2/23/17, Bhupesh Rawat <[hidden email]> wrote:
>> Thank you so much for the response!!
>> On Feb 23, 2017 8:26 AM, "Eibe Frank" <[hidden email]> wrote:
>>
>>> You mean beyond confidence, lift, or one the other metrics that you can
>>> get in the output of each rule? This is a tough question. One way may be
>>> to
>>> use the association rule mining algorithm to build classification rules
>>> and
>>> then evaluate the accuracy of those classification rules. We had a paper
>>> on
>>> this quite a while back:
>>>
>>> Mutter, S., Hall, M., & Frank, E. (2004, December). Using classification
>>> to evaluate the output of confidence-based association rule mining. In
>>> Australasian Joint Conference on Artificial Intelligence (pp. 538-549).
>>> Springer Berlin Heidelberg.
>>>
>>> I suppose you could also evaluate the individual association rules on a
>>> separate test set, by computing the confidence measure, etc., on the test
>>> set for each rule, but this functionality is not provided by WEKA.
>>>
>>> Cheers,
>>> Eibe
>>>
>>>> On 23/02/2017, at 12:46 AM, Bhupesh Rawat <[hidden email]> wrote:
>>>>
>>>> Dear Sir/Madam
>>>>
>>>> I have discovered some rules through weka. Could you tell me how to
>>> measure  the accuracy of those rules.
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> On Tue, Feb 14, 2017 at 3:44 AM, Peter Reutemann
>>>> <[hidden email]>
>>> wrote:
>>>>> The size of the final Apriori model as a serialised Java object can
>>>>> be
>>> established saving it to a file and considering the file size. Note that
>>> this is different from the size of the object in memory (see, e.g.,
>>> http://stackoverflow.com/questions/7146559/serialized-
>>> object-size-vs-in-memory-object-size-in-java#7146941).
>>>>>
>>>>> I don’t know of a good way to measure peak memory consumption of a
>>> Java program (after garbage collection). A crude way would be to run the
>>> program from the command-line (to avoid overhead associated with the
>>> GUIs)
>>> with different maximum heap sizes, e.g., increasing the heap size until
>>> the
>>> program runs through. Another option is to look at the heap size in a
>>> profiler (e.g., visualvm), enforcing garbage collection before a readout.
>>>>
>>>> You can use the sizeofag javaagent for determining the size of a Java
>>> object:
>>>> https://github.com/fracpete/sizeofag
>>>>
>>>> Credits to Maxim Zakharenkov, who wrote the original code.
>>>>
>>>> Cheers, Peter
>>>> --
>>>> Peter Reutemann
>>>> Dept. of Computer Science
>>>> University of Waikato, NZ
>>>> +64 (7) 858-5174
>>>> http://www.cms.waikato.ac.nz/~fracpete/
>>>> http://www.data-mining.co.nz/
>>>> _______________________________________________
>>>> Wekalist mailing list
>>>> Send posts to: [hidden email]
>>>> List info and subscription status: https://list.waikato.ac.nz/
>>> mailman/listinfo/wekalist
>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>> ml/weka/mailinglist_etiquette.html
>>>>
>>>>
>>>>
>>>> --
>>>> Thanks & Regards
>>>> Bhupesh Rawat.
>>>> Ph.D Scholar
>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>> Ph. No: +91-9897065948
>>>>
>>>> ............................................................
>>> ...............................................................
>>>> *A man is the best judge of himself and he has to pay the price for
>>>> what
>>> he
>>>> does.*
>>>> ............................................................
>>> ...............................................................
>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> Wekalist mailing list
>>>> Send posts to: [hidden email]
>>>> List info and subscription status: https://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: [hidden email]
>>> List info and subscription status: https://list.waikato.ac.nz/
>>> mailman/listinfo/wekalist
>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>> ml/weka/mailinglist_etiquette.html
>>>
>>
>
>
> --
> Thanks & Regards
> Bhupesh Rawat.
> Ph.D Scholar
> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
> Vidya Vihar,Rai Bareilly road(Lucknow)
> Ph. No: +91-9897065948
>
> ...........................................................................................................................
> *A man is the best judge of himself and he has to pay the price for what he
> does.*
> ...........................................................................................................................
> <students' data after preprocessin.xlsx>_______________________________________________
> Wekalist mailing list
> Send posts to: [hidden email]
> List info and subscription status: https://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: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Bhupesh Rawat
Sir,

How could i perform these two task seperately(applying classification
rule learner and estimating classification accuracy). The accuracy is
estimated each time i run the classifier on the dataset.



On 2/24/17, Eibe Frank <[hidden email]> wrote:

> No, not really. However, the dataset is quite small. You could just run a
> classification rule learner such as JRip or PART on the data, treating each
> of the attributes in turn as the class attribute. Then you can estimate
> classification accuracy using cross-validation.
>
> You could also create combinations of attributes using the CartesionProduct
> filter.
>
> Cheers,
> Eibe
>
>> On 24/02/2017, at 3:11 AM, Bhupesh Rawat <[hidden email]> wrote:
>>
>> I have a small dataset which contains student enrolment data in
>> various courses. If a student has selected a particular course it is
>> indicated by ‘Y’ else ‘N’ is used. I have also attached a file for
>> better understanding of the dataset. I am interested in knowing if it
>> is possible to measure the accuracy of the association rules with this
>> dataset by the proposed approach in your paper.
>>
>> On 2/23/17, Bhupesh Rawat <[hidden email]> wrote:
>>> Thank you so much for the response!!
>>> On Feb 23, 2017 8:26 AM, "Eibe Frank" <[hidden email]> wrote:
>>>
>>>> You mean beyond confidence, lift, or one the other metrics that you can
>>>> get in the output of each rule? This is a tough question. One way may be
>>>> to
>>>> use the association rule mining algorithm to build classification rules
>>>> and
>>>> then evaluate the accuracy of those classification rules. We had a paper
>>>> on
>>>> this quite a while back:
>>>>
>>>> Mutter, S., Hall, M., & Frank, E. (2004, December). Using classification
>>>> to evaluate the output of confidence-based association rule mining. In
>>>> Australasian Joint Conference on Artificial Intelligence (pp. 538-549).
>>>> Springer Berlin Heidelberg.
>>>>
>>>> I suppose you could also evaluate the individual association rules on a
>>>> separate test set, by computing the confidence measure, etc., on the
>>>> test
>>>> set for each rule, but this functionality is not provided by WEKA.
>>>>
>>>> Cheers,
>>>> Eibe
>>>>
>>>>> On 23/02/2017, at 12:46 AM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>
>>>>> Dear Sir/Madam
>>>>>
>>>>> I have discovered some rules through weka. Could you tell me how to
>>>> measure  the accuracy of those rules.
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> On Tue, Feb 14, 2017 at 3:44 AM, Peter Reutemann
>>>>> <[hidden email]>
>>>> wrote:
>>>>>> The size of the final Apriori model as a serialised Java object can
>>>>>> be
>>>> established saving it to a file and considering the file size. Note that
>>>> this is different from the size of the object in memory (see, e.g.,
>>>> http://stackoverflow.com/questions/7146559/serialized-
>>>> object-size-vs-in-memory-object-size-in-java#7146941).
>>>>>>
>>>>>> I don’t know of a good way to measure peak memory consumption of a
>>>> Java program (after garbage collection). A crude way would be to run the
>>>> program from the command-line (to avoid overhead associated with the
>>>> GUIs)
>>>> with different maximum heap sizes, e.g., increasing the heap size until
>>>> the
>>>> program runs through. Another option is to look at the heap size in a
>>>> profiler (e.g., visualvm), enforcing garbage collection before a
>>>> readout.
>>>>>
>>>>> You can use the sizeofag javaagent for determining the size of a Java
>>>> object:
>>>>> https://github.com/fracpete/sizeofag
>>>>>
>>>>> Credits to Maxim Zakharenkov, who wrote the original code.
>>>>>
>>>>> Cheers, Peter
>>>>> --
>>>>> Peter Reutemann
>>>>> Dept. of Computer Science
>>>>> University of Waikato, NZ
>>>>> +64 (7) 858-5174
>>>>> http://www.cms.waikato.ac.nz/~fracpete/
>>>>> http://www.data-mining.co.nz/
>>>>> _______________________________________________
>>>>> Wekalist mailing list
>>>>> Send posts to: [hidden email]
>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>> mailman/listinfo/wekalist
>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>> ml/weka/mailinglist_etiquette.html
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> Thanks & Regards
>>>>> Bhupesh Rawat.
>>>>> Ph.D Scholar
>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>> Ph. No: +91-9897065948
>>>>>
>>>>> ............................................................
>>>> ...............................................................
>>>>> *A man is the best judge of himself and he has to pay the price for
>>>>> what
>>>> he
>>>>> does.*
>>>>> ............................................................
>>>> ...............................................................
>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> Wekalist mailing list
>>>>> Send posts to: [hidden email]
>>>>> List info and subscription status: https://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: [hidden email]
>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>> mailman/listinfo/wekalist
>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>> ml/weka/mailinglist_etiquette.html
>>>>
>>>
>>
>>
>> --
>> Thanks & Regards
>> Bhupesh Rawat.
>> Ph.D Scholar
>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>> Vidya Vihar,Rai Bareilly road(Lucknow)
>> Ph. No: +91-9897065948
>>
>> ...........................................................................................................................
>> *A man is the best judge of himself and he has to pay the price for what
>> he
>> does.*
>> ...........................................................................................................................
>> <students' data after
>> preprocessin.xlsx>_______________________________________________
>> Wekalist mailing list
>> Send posts to: [hidden email]
>> List info and subscription status:
>> https://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: [hidden email]
> List info and subscription status:
> https://list.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>


--
Thanks & Regards
Bhupesh Rawat.
Ph.D Scholar
Department of Computer Science,Babasaheb Bhimrao Ambedkar University
Vidya Vihar,Rai Bareilly road(Lucknow)
Ph. No: +91-9897065948

...........................................................................................................................
*A man is the best judge of himself and he has to pay the price for what he
does.*
...........................................................................................................................
_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Eibe Frank-2
Administrator
In the Explorer, there is no way to turn off evaluation completely. You could use the command-line interface or the KnowledgeFlow GUI though.

Having said this, if you evaluate on the training set, the runtime overhead is quite small if you apply a rule learner.

Note also that the Explorer always outputs the classification model for the *full* dataset loaded into the Preprocess panel, regardless of which evaluation metric you choose, i.e., you will get the rule set for the full dataset regardless of the evaluation method you use.

Cheers,
Eibe

> On 26/02/2017, at 8:07 PM, Bhupesh Rawat <[hidden email]> wrote:
>
> Sir,
>
> How could i perform these two task seperately(applying classification
> rule learner and estimating classification accuracy). The accuracy is
> estimated each time i run the classifier on the dataset.
>
>
>
> On 2/24/17, Eibe Frank <[hidden email]> wrote:
>> No, not really. However, the dataset is quite small. You could just run a
>> classification rule learner such as JRip or PART on the data, treating each
>> of the attributes in turn as the class attribute. Then you can estimate
>> classification accuracy using cross-validation.
>>
>> You could also create combinations of attributes using the CartesionProduct
>> filter.
>>
>> Cheers,
>> Eibe
>>
>>> On 24/02/2017, at 3:11 AM, Bhupesh Rawat <[hidden email]> wrote:
>>>
>>> I have a small dataset which contains student enrolment data in
>>> various courses. If a student has selected a particular course it is
>>> indicated by ‘Y’ else ‘N’ is used. I have also attached a file for
>>> better understanding of the dataset. I am interested in knowing if it
>>> is possible to measure the accuracy of the association rules with this
>>> dataset by the proposed approach in your paper.
>>>
>>> On 2/23/17, Bhupesh Rawat <[hidden email]> wrote:
>>>> Thank you so much for the response!!
>>>> On Feb 23, 2017 8:26 AM, "Eibe Frank" <[hidden email]> wrote:
>>>>
>>>>> You mean beyond confidence, lift, or one the other metrics that you can
>>>>> get in the output of each rule? This is a tough question. One way may be
>>>>> to
>>>>> use the association rule mining algorithm to build classification rules
>>>>> and
>>>>> then evaluate the accuracy of those classification rules. We had a paper
>>>>> on
>>>>> this quite a while back:
>>>>>
>>>>> Mutter, S., Hall, M., & Frank, E. (2004, December). Using classification
>>>>> to evaluate the output of confidence-based association rule mining. In
>>>>> Australasian Joint Conference on Artificial Intelligence (pp. 538-549).
>>>>> Springer Berlin Heidelberg.
>>>>>
>>>>> I suppose you could also evaluate the individual association rules on a
>>>>> separate test set, by computing the confidence measure, etc., on the
>>>>> test
>>>>> set for each rule, but this functionality is not provided by WEKA.
>>>>>
>>>>> Cheers,
>>>>> Eibe
>>>>>
>>>>>> On 23/02/2017, at 12:46 AM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>
>>>>>> Dear Sir/Madam
>>>>>>
>>>>>> I have discovered some rules through weka. Could you tell me how to
>>>>> measure  the accuracy of those rules.
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> On Tue, Feb 14, 2017 at 3:44 AM, Peter Reutemann
>>>>>> <[hidden email]>
>>>>> wrote:
>>>>>>> The size of the final Apriori model as a serialised Java object can
>>>>>>> be
>>>>> established saving it to a file and considering the file size. Note that
>>>>> this is different from the size of the object in memory (see, e.g.,
>>>>> http://stackoverflow.com/questions/7146559/serialized-
>>>>> object-size-vs-in-memory-object-size-in-java#7146941).
>>>>>>>
>>>>>>> I don’t know of a good way to measure peak memory consumption of a
>>>>> Java program (after garbage collection). A crude way would be to run the
>>>>> program from the command-line (to avoid overhead associated with the
>>>>> GUIs)
>>>>> with different maximum heap sizes, e.g., increasing the heap size until
>>>>> the
>>>>> program runs through. Another option is to look at the heap size in a
>>>>> profiler (e.g., visualvm), enforcing garbage collection before a
>>>>> readout.
>>>>>>
>>>>>> You can use the sizeofag javaagent for determining the size of a Java
>>>>> object:
>>>>>> https://github.com/fracpete/sizeofag
>>>>>>
>>>>>> Credits to Maxim Zakharenkov, who wrote the original code.
>>>>>>
>>>>>> Cheers, Peter
>>>>>> --
>>>>>> Peter Reutemann
>>>>>> Dept. of Computer Science
>>>>>> University of Waikato, NZ
>>>>>> +64 (7) 858-5174
>>>>>> http://www.cms.waikato.ac.nz/~fracpete/
>>>>>> http://www.data-mining.co.nz/
>>>>>> _______________________________________________
>>>>>> Wekalist mailing list
>>>>>> Send posts to: [hidden email]
>>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>>> mailman/listinfo/wekalist
>>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>>> ml/weka/mailinglist_etiquette.html
>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> Thanks & Regards
>>>>>> Bhupesh Rawat.
>>>>>> Ph.D Scholar
>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>> Ph. No: +91-9897065948
>>>>>>
>>>>>> ............................................................
>>>>> ...............................................................
>>>>>> *A man is the best judge of himself and he has to pay the price for
>>>>>> what
>>>>> he
>>>>>> does.*
>>>>>> ............................................................
>>>>> ...............................................................
>>>>>>
>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> Wekalist mailing list
>>>>>> Send posts to: [hidden email]
>>>>>> List info and subscription status: https://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: [hidden email]
>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>>> mailman/listinfo/wekalist
>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>>> ml/weka/mailinglist_etiquette.html
>>>>>
>>>>
>>>
>>>
>>> --
>>> Thanks & Regards
>>> Bhupesh Rawat.
>>> Ph.D Scholar
>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>> Ph. No: +91-9897065948
>>>
>>> ...........................................................................................................................
>>> *A man is the best judge of himself and he has to pay the price for what
>>> he
>>> does.*
>>> ...........................................................................................................................
>>> <students' data after
>>> preprocessin.xlsx>_______________________________________________
>>> Wekalist mailing list
>>> Send posts to: [hidden email]
>>> List info and subscription status:
>>> https://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: [hidden email]
>> List info and subscription status:
>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>> List etiquette:
>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>
>
>
> --
> Thanks & Regards
> Bhupesh Rawat.
> Ph.D Scholar
> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
> Vidya Vihar,Rai Bareilly road(Lucknow)
> Ph. No: +91-9897065948
>
> ...........................................................................................................................
> *A man is the best judge of himself and he has to pay the price for what he
> does.*
> ...........................................................................................................................
> _______________________________________________
> Wekalist mailing list
> Send posts to: [hidden email]
> List info and subscription status: https://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: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Bhupesh Rawat
Sir,
When I use the KnowledgeFlow GUI the status shown by two of the components is interrupted(namely crossvalidationfoldmaker and J48) as shown in the attached file. How to fix it? 

On Mon, Feb 27, 2017 at 3:08 AM, Eibe Frank <[hidden email]> wrote:
In the Explorer, there is no way to turn off evaluation completely. You could use the command-line interface or the KnowledgeFlow GUI though.

Having said this, if you evaluate on the training set, the runtime overhead is quite small if you apply a rule learner.

Note also that the Explorer always outputs the classification model for the *full* dataset loaded into the Preprocess panel, regardless of which evaluation metric you choose, i.e., you will get the rule set for the full dataset regardless of the evaluation method you use.

Cheers,
Eibe

> On 26/02/2017, at 8:07 PM, Bhupesh Rawat <[hidden email]> wrote:
>
> Sir,
>
> How could i perform these two task seperately(applying classification
> rule learner and estimating classification accuracy). The accuracy is
> estimated each time i run the classifier on the dataset.
>
>
>
> On 2/24/17, Eibe Frank <[hidden email]> wrote:
>> No, not really. However, the dataset is quite small. You could just run a
>> classification rule learner such as JRip or PART on the data, treating each
>> of the attributes in turn as the class attribute. Then you can estimate
>> classification accuracy using cross-validation.
>>
>> You could also create combinations of attributes using the CartesionProduct
>> filter.
>>
>> Cheers,
>> Eibe
>>
>>> On 24/02/2017, at 3:11 AM, Bhupesh Rawat <[hidden email]> wrote:
>>>
>>> I have a small dataset which contains student enrolment data in
>>> various courses. If a student has selected a particular course it is
>>> indicated by ‘Y’ else ‘N’ is used. I have also attached a file for
>>> better understanding of the dataset. I am interested in knowing if it
>>> is possible to measure the accuracy of the association rules with this
>>> dataset by the proposed approach in your paper.
>>>
>>> On 2/23/17, Bhupesh Rawat <[hidden email]> wrote:
>>>> Thank you so much for the response!!
>>>> On Feb 23, 2017 8:26 AM, "Eibe Frank" <[hidden email]> wrote:
>>>>
>>>>> You mean beyond confidence, lift, or one the other metrics that you can
>>>>> get in the output of each rule? This is a tough question. One way may be
>>>>> to
>>>>> use the association rule mining algorithm to build classification rules
>>>>> and
>>>>> then evaluate the accuracy of those classification rules. We had a paper
>>>>> on
>>>>> this quite a while back:
>>>>>
>>>>> Mutter, S., Hall, M., & Frank, E. (2004, December). Using classification
>>>>> to evaluate the output of confidence-based association rule mining. In
>>>>> Australasian Joint Conference on Artificial Intelligence (pp. 538-549).
>>>>> Springer Berlin Heidelberg.
>>>>>
>>>>> I suppose you could also evaluate the individual association rules on a
>>>>> separate test set, by computing the confidence measure, etc., on the
>>>>> test
>>>>> set for each rule, but this functionality is not provided by WEKA.
>>>>>
>>>>> Cheers,
>>>>> Eibe
>>>>>
>>>>>> On 23/02/2017, at 12:46 AM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>
>>>>>> Dear Sir/Madam
>>>>>>
>>>>>> I have discovered some rules through weka. Could you tell me how to
>>>>> measure  the accuracy of those rules.
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> On Tue, Feb 14, 2017 at 3:44 AM, Peter Reutemann
>>>>>> <[hidden email]>
>>>>> wrote:
>>>>>>> The size of the final Apriori model as a serialised Java object can
>>>>>>> be
>>>>> established saving it to a file and considering the file size. Note that
>>>>> this is different from the size of the object in memory (see, e.g.,
>>>>> http://stackoverflow.com/questions/7146559/serialized-
>>>>> object-size-vs-in-memory-object-size-in-java#7146941).
>>>>>>>
>>>>>>> I don’t know of a good way to measure peak memory consumption of a
>>>>> Java program (after garbage collection). A crude way would be to run the
>>>>> program from the command-line (to avoid overhead associated with the
>>>>> GUIs)
>>>>> with different maximum heap sizes, e.g., increasing the heap size until
>>>>> the
>>>>> program runs through. Another option is to look at the heap size in a
>>>>> profiler (e.g., visualvm), enforcing garbage collection before a
>>>>> readout.
>>>>>>
>>>>>> You can use the sizeofag javaagent for determining the size of a Java
>>>>> object:
>>>>>> https://github.com/fracpete/sizeofag
>>>>>>
>>>>>> Credits to Maxim Zakharenkov, who wrote the original code.
>>>>>>
>>>>>> Cheers, Peter
>>>>>> --
>>>>>> Peter Reutemann
>>>>>> Dept. of Computer Science
>>>>>> University of Waikato, NZ
>>>>>> +64 (7) 858-5174
>>>>>> http://www.cms.waikato.ac.nz/~fracpete/
>>>>>> http://www.data-mining.co.nz/
>>>>>> _______________________________________________
>>>>>> Wekalist mailing list
>>>>>> Send posts to: [hidden email]
>>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>>> mailman/listinfo/wekalist
>>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>>> ml/weka/mailinglist_etiquette.html
>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> Thanks & Regards
>>>>>> Bhupesh Rawat.
>>>>>> Ph.D Scholar
>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>> Ph. No: +91-9897065948
>>>>>>
>>>>>> ............................................................
>>>>> ...............................................................
>>>>>> *A man is the best judge of himself and he has to pay the price for
>>>>>> what
>>>>> he
>>>>>> does.*
>>>>>> ............................................................
>>>>> ...............................................................
>>>>>>
>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> Wekalist mailing list
>>>>>> Send posts to: [hidden email]
>>>>>> List info and subscription status: https://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: [hidden email]
>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>>> mailman/listinfo/wekalist
>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>>> ml/weka/mailinglist_etiquette.html
>>>>>
>>>>
>>>
>>>
>>> --
>>> Thanks & Regards
>>> Bhupesh Rawat.
>>> Ph.D Scholar
>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>> Ph. No: +91-9897065948
>>>
>>> ...........................................................................................................................
>>> *A man is the best judge of himself and he has to pay the price for what
>>> he
>>> does.*
>>> ...........................................................................................................................
>>> <students' data after
>>> preprocessin.xlsx>_______________________________________________
>>> Wekalist mailing list
>>> Send posts to: [hidden email]
>>> List info and subscription status:
>>> https://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: [hidden email]
>> List info and subscription status:
>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>> List etiquette:
>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>
>
>
> --
> Thanks & Regards
> Bhupesh Rawat.
> Ph.D Scholar
> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
> Vidya Vihar,Rai Bareilly road(Lucknow)
> Ph. No: +91-9897065948
>
> ...........................................................................................................................
> *A man is the best judge of himself and he has to pay the price for what he
> does.*
> ...........................................................................................................................
> _______________________________________________
> Wekalist mailing list
> Send posts to: [hidden email]
> List info and subscription status: https://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: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html



--
Thanks & Regards
Bhupesh Rawat.
Ph.D Scholar
Department of Computer Science,Babasaheb Bhimrao Ambedkar University
Vidya Vihar,Rai Bareilly road(Lucknow)
Ph. No: +91-9897065948

...........................................................................................................................
*A man is the best judge of himself and he has to pay the price for what he
does.*
...........................................................................................................................




_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html

knowledge flow interuppted.docx (248K) Download Attachment
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Eibe Frank-2
Administrator
What does the log say (see the “log” tab next to the “status” tab)?

Cheers,
Eibe

> On 27/02/2017, at 11:56 PM, Bhupesh Rawat <[hidden email]> wrote:
>
> Sir,
> When I use the KnowledgeFlow GUI the status shown by two of the components is interrupted(namely crossvalidationfoldmaker and J48) as shown in the attached file. How to fix it?
>
> On Mon, Feb 27, 2017 at 3:08 AM, Eibe Frank <[hidden email]> wrote:
> In the Explorer, there is no way to turn off evaluation completely. You could use the command-line interface or the KnowledgeFlow GUI though.
>
> Having said this, if you evaluate on the training set, the runtime overhead is quite small if you apply a rule learner.
>
> Note also that the Explorer always outputs the classification model for the *full* dataset loaded into the Preprocess panel, regardless of which evaluation metric you choose, i.e., you will get the rule set for the full dataset regardless of the evaluation method you use.
>
> Cheers,
> Eibe
>
> > On 26/02/2017, at 8:07 PM, Bhupesh Rawat <[hidden email]> wrote:
> >
> > Sir,
> >
> > How could i perform these two task seperately(applying classification
> > rule learner and estimating classification accuracy). The accuracy is
> > estimated each time i run the classifier on the dataset.
> >
> >
> >
> > On 2/24/17, Eibe Frank <[hidden email]> wrote:
> >> No, not really. However, the dataset is quite small. You could just run a
> >> classification rule learner such as JRip or PART on the data, treating each
> >> of the attributes in turn as the class attribute. Then you can estimate
> >> classification accuracy using cross-validation.
> >>
> >> You could also create combinations of attributes using the CartesionProduct
> >> filter.
> >>
> >> Cheers,
> >> Eibe
> >>
> >>> On 24/02/2017, at 3:11 AM, Bhupesh Rawat <[hidden email]> wrote:
> >>>
> >>> I have a small dataset which contains student enrolment data in
> >>> various courses. If a student has selected a particular course it is
> >>> indicated by ‘Y’ else ‘N’ is used. I have also attached a file for
> >>> better understanding of the dataset. I am interested in knowing if it
> >>> is possible to measure the accuracy of the association rules with this
> >>> dataset by the proposed approach in your paper.
> >>>
> >>> On 2/23/17, Bhupesh Rawat <[hidden email]> wrote:
> >>>> Thank you so much for the response!!
> >>>> On Feb 23, 2017 8:26 AM, "Eibe Frank" <[hidden email]> wrote:
> >>>>
> >>>>> You mean beyond confidence, lift, or one the other metrics that you can
> >>>>> get in the output of each rule? This is a tough question. One way may be
> >>>>> to
> >>>>> use the association rule mining algorithm to build classification rules
> >>>>> and
> >>>>> then evaluate the accuracy of those classification rules. We had a paper
> >>>>> on
> >>>>> this quite a while back:
> >>>>>
> >>>>> Mutter, S., Hall, M., & Frank, E. (2004, December). Using classification
> >>>>> to evaluate the output of confidence-based association rule mining. In
> >>>>> Australasian Joint Conference on Artificial Intelligence (pp. 538-549).
> >>>>> Springer Berlin Heidelberg.
> >>>>>
> >>>>> I suppose you could also evaluate the individual association rules on a
> >>>>> separate test set, by computing the confidence measure, etc., on the
> >>>>> test
> >>>>> set for each rule, but this functionality is not provided by WEKA.
> >>>>>
> >>>>> Cheers,
> >>>>> Eibe
> >>>>>
> >>>>>> On 23/02/2017, at 12:46 AM, Bhupesh Rawat <[hidden email]> wrote:
> >>>>>>
> >>>>>> Dear Sir/Madam
> >>>>>>
> >>>>>> I have discovered some rules through weka. Could you tell me how to
> >>>>> measure  the accuracy of those rules.
> >>>>>>
> >>>>>>
> >>>>>>
> >>>>>>
> >>>>>>
> >>>>>>
> >>>>>>
> >>>>>> On Tue, Feb 14, 2017 at 3:44 AM, Peter Reutemann
> >>>>>> <[hidden email]>
> >>>>> wrote:
> >>>>>>> The size of the final Apriori model as a serialised Java object can
> >>>>>>> be
> >>>>> established saving it to a file and considering the file size. Note that
> >>>>> this is different from the size of the object in memory (see, e.g.,
> >>>>> http://stackoverflow.com/questions/7146559/serialized-
> >>>>> object-size-vs-in-memory-object-size-in-java#7146941).
> >>>>>>>
> >>>>>>> I don’t know of a good way to measure peak memory consumption of a
> >>>>> Java program (after garbage collection). A crude way would be to run the
> >>>>> program from the command-line (to avoid overhead associated with the
> >>>>> GUIs)
> >>>>> with different maximum heap sizes, e.g., increasing the heap size until
> >>>>> the
> >>>>> program runs through. Another option is to look at the heap size in a
> >>>>> profiler (e.g., visualvm), enforcing garbage collection before a
> >>>>> readout.
> >>>>>>
> >>>>>> You can use the sizeofag javaagent for determining the size of a Java
> >>>>> object:
> >>>>>> https://github.com/fracpete/sizeofag
> >>>>>>
> >>>>>> Credits to Maxim Zakharenkov, who wrote the original code.
> >>>>>>
> >>>>>> Cheers, Peter
> >>>>>> --
> >>>>>> Peter Reutemann
> >>>>>> Dept. of Computer Science
> >>>>>> University of Waikato, NZ
> >>>>>> +64 (7) 858-5174
> >>>>>> http://www.cms.waikato.ac.nz/~fracpete/
> >>>>>> http://www.data-mining.co.nz/
> >>>>>> _______________________________________________
> >>>>>> Wekalist mailing list
> >>>>>> Send posts to: [hidden email]
> >>>>>> List info and subscription status: https://list.waikato.ac.nz/
> >>>>> mailman/listinfo/wekalist
> >>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
> >>>>> ml/weka/mailinglist_etiquette.html
> >>>>>>
> >>>>>>
> >>>>>>
> >>>>>> --
> >>>>>> Thanks & Regards
> >>>>>> Bhupesh Rawat.
> >>>>>> Ph.D Scholar
> >>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
> >>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
> >>>>>> Ph. No: +91-9897065948
> >>>>>>
> >>>>>> ............................................................
> >>>>> ...............................................................
> >>>>>> *A man is the best judge of himself and he has to pay the price for
> >>>>>> what
> >>>>> he
> >>>>>> does.*
> >>>>>> ............................................................
> >>>>> ...............................................................
> >>>>>>
> >>>>>>
> >>>>>>
> >>>>>> _______________________________________________
> >>>>>> Wekalist mailing list
> >>>>>> Send posts to: [hidden email]
> >>>>>> List info and subscription status: https://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: [hidden email]
> >>>>> List info and subscription status: https://list.waikato.ac.nz/
> >>>>> mailman/listinfo/wekalist
> >>>>> List etiquette: http://www.cs.waikato.ac.nz/~
> >>>>> ml/weka/mailinglist_etiquette.html
> >>>>>
> >>>>
> >>>
> >>>
> >>> --
> >>> Thanks & Regards
> >>> Bhupesh Rawat.
> >>> Ph.D Scholar
> >>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
> >>> Vidya Vihar,Rai Bareilly road(Lucknow)
> >>> Ph. No: +91-9897065948
> >>>
> >>> ...........................................................................................................................
> >>> *A man is the best judge of himself and he has to pay the price for what
> >>> he
> >>> does.*
> >>> ...........................................................................................................................
> >>> <students' data after
> >>> preprocessin.xlsx>_______________________________________________
> >>> Wekalist mailing list
> >>> Send posts to: [hidden email]
> >>> List info and subscription status:
> >>> https://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: [hidden email]
> >> List info and subscription status:
> >> https://list.waikato.ac.nz/mailman/listinfo/wekalist
> >> List etiquette:
> >> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> >>
> >
> >
> > --
> > Thanks & Regards
> > Bhupesh Rawat.
> > Ph.D Scholar
> > Department of Computer Science,Babasaheb Bhimrao Ambedkar University
> > Vidya Vihar,Rai Bareilly road(Lucknow)
> > Ph. No: +91-9897065948
> >
> > ...........................................................................................................................
> > *A man is the best judge of himself and he has to pay the price for what he
> > does.*
> > ...........................................................................................................................
> > _______________________________________________
> > Wekalist mailing list
> > Send posts to: [hidden email]
> > List info and subscription status: https://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: [hidden email]
> List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>
>
>
> --
> Thanks & Regards
> Bhupesh Rawat.
> Ph.D Scholar
> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
> Vidya Vihar,Rai Bareilly road(Lucknow)
> Ph. No: +91-9897065948
>
> ...........................................................................................................................
> *A man is the best judge of himself and he has to pay the price for what he
> does.*
> ...........................................................................................................................
>
>
>
> <knowledge flow  interuppted.docx>_______________________________________________
> Wekalist mailing list
> Send posts to: [hidden email]
> List info and subscription status: https://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: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Bhupesh Rawat
Thank you Sir, the problem has been fixed.

Moreover i would also like to use the combination of attributes for
which  you suggested  the CartesionProduct filter. Where could i find
this option?

On 2/28/17, Eibe Frank <[hidden email]> wrote:

> What does the log say (see the “log” tab next to the “status” tab)?
>
> Cheers,
> Eibe
>
>> On 27/02/2017, at 11:56 PM, Bhupesh Rawat <[hidden email]> wrote:
>>
>> Sir,
>> When I use the KnowledgeFlow GUI the status shown by two of the components
>> is interrupted(namely crossvalidationfoldmaker and J48) as shown in the
>> attached file. How to fix it?
>>
>> On Mon, Feb 27, 2017 at 3:08 AM, Eibe Frank <[hidden email]> wrote:
>> In the Explorer, there is no way to turn off evaluation completely. You
>> could use the command-line interface or the KnowledgeFlow GUI though.
>>
>> Having said this, if you evaluate on the training set, the runtime
>> overhead is quite small if you apply a rule learner.
>>
>> Note also that the Explorer always outputs the classification model for
>> the *full* dataset loaded into the Preprocess panel, regardless of which
>> evaluation metric you choose, i.e., you will get the rule set for the full
>> dataset regardless of the evaluation method you use.
>>
>> Cheers,
>> Eibe
>>
>> > On 26/02/2017, at 8:07 PM, Bhupesh Rawat <[hidden email]> wrote:
>> >
>> > Sir,
>> >
>> > How could i perform these two task seperately(applying classification
>> > rule learner and estimating classification accuracy). The accuracy is
>> > estimated each time i run the classifier on the dataset.
>> >
>> >
>> >
>> > On 2/24/17, Eibe Frank <[hidden email]> wrote:
>> >> No, not really. However, the dataset is quite small. You could just run
>> >> a
>> >> classification rule learner such as JRip or PART on the data, treating
>> >> each
>> >> of the attributes in turn as the class attribute. Then you can estimate
>> >> classification accuracy using cross-validation.
>> >>
>> >> You could also create combinations of attributes using the
>> >> CartesionProduct
>> >> filter.
>> >>
>> >> Cheers,
>> >> Eibe
>> >>
>> >>> On 24/02/2017, at 3:11 AM, Bhupesh Rawat <[hidden email]> wrote:
>> >>>
>> >>> I have a small dataset which contains student enrolment data in
>> >>> various courses. If a student has selected a particular course it is
>> >>> indicated by ‘Y’ else ‘N’ is used. I have also attached a file for
>> >>> better understanding of the dataset. I am interested in knowing if it
>> >>> is possible to measure the accuracy of the association rules with this
>> >>> dataset by the proposed approach in your paper.
>> >>>
>> >>> On 2/23/17, Bhupesh Rawat <[hidden email]> wrote:
>> >>>> Thank you so much for the response!!
>> >>>> On Feb 23, 2017 8:26 AM, "Eibe Frank" <[hidden email]> wrote:
>> >>>>
>> >>>>> You mean beyond confidence, lift, or one the other metrics that you
>> >>>>> can
>> >>>>> get in the output of each rule? This is a tough question. One way
>> >>>>> may be
>> >>>>> to
>> >>>>> use the association rule mining algorithm to build classification
>> >>>>> rules
>> >>>>> and
>> >>>>> then evaluate the accuracy of those classification rules. We had a
>> >>>>> paper
>> >>>>> on
>> >>>>> this quite a while back:
>> >>>>>
>> >>>>> Mutter, S., Hall, M., & Frank, E. (2004, December). Using
>> >>>>> classification
>> >>>>> to evaluate the output of confidence-based association rule mining.
>> >>>>> In
>> >>>>> Australasian Joint Conference on Artificial Intelligence (pp.
>> >>>>> 538-549).
>> >>>>> Springer Berlin Heidelberg.
>> >>>>>
>> >>>>> I suppose you could also evaluate the individual association rules
>> >>>>> on a
>> >>>>> separate test set, by computing the confidence measure, etc., on the
>> >>>>> test
>> >>>>> set for each rule, but this functionality is not provided by WEKA.
>> >>>>>
>> >>>>> Cheers,
>> >>>>> Eibe
>> >>>>>
>> >>>>>> On 23/02/2017, at 12:46 AM, Bhupesh Rawat <[hidden email]> wrote:
>> >>>>>>
>> >>>>>> Dear Sir/Madam
>> >>>>>>
>> >>>>>> I have discovered some rules through weka. Could you tell me how to
>> >>>>> measure  the accuracy of those rules.
>> >>>>>>
>> >>>>>>
>> >>>>>>
>> >>>>>>
>> >>>>>>
>> >>>>>>
>> >>>>>>
>> >>>>>> On Tue, Feb 14, 2017 at 3:44 AM, Peter Reutemann
>> >>>>>> <[hidden email]>
>> >>>>> wrote:
>> >>>>>>> The size of the final Apriori model as a serialised Java object
>> >>>>>>> can
>> >>>>>>> be
>> >>>>> established saving it to a file and considering the file size. Note
>> >>>>> that
>> >>>>> this is different from the size of the object in memory (see, e.g.,
>> >>>>> http://stackoverflow.com/questions/7146559/serialized-
>> >>>>> object-size-vs-in-memory-object-size-in-java#7146941).
>> >>>>>>>
>> >>>>>>> I don’t know of a good way to measure peak memory consumption of a
>> >>>>> Java program (after garbage collection). A crude way would be to run
>> >>>>> the
>> >>>>> program from the command-line (to avoid overhead associated with the
>> >>>>> GUIs)
>> >>>>> with different maximum heap sizes, e.g., increasing the heap size
>> >>>>> until
>> >>>>> the
>> >>>>> program runs through. Another option is to look at the heap size in
>> >>>>> a
>> >>>>> profiler (e.g., visualvm), enforcing garbage collection before a
>> >>>>> readout.
>> >>>>>>
>> >>>>>> You can use the sizeofag javaagent for determining the size of a
>> >>>>>> Java
>> >>>>> object:
>> >>>>>> https://github.com/fracpete/sizeofag
>> >>>>>>
>> >>>>>> Credits to Maxim Zakharenkov, who wrote the original code.
>> >>>>>>
>> >>>>>> Cheers, Peter
>> >>>>>> --
>> >>>>>> Peter Reutemann
>> >>>>>> Dept. of Computer Science
>> >>>>>> University of Waikato, NZ
>> >>>>>> +64 (7) 858-5174
>> >>>>>> http://www.cms.waikato.ac.nz/~fracpete/
>> >>>>>> http://www.data-mining.co.nz/
>> >>>>>> _______________________________________________
>> >>>>>> Wekalist mailing list
>> >>>>>> Send posts to: [hidden email]
>> >>>>>> List info and subscription status: https://list.waikato.ac.nz/
>> >>>>> mailman/listinfo/wekalist
>> >>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>> >>>>> ml/weka/mailinglist_etiquette.html
>> >>>>>>
>> >>>>>>
>> >>>>>>
>> >>>>>> --
>> >>>>>> Thanks & Regards
>> >>>>>> Bhupesh Rawat.
>> >>>>>> Ph.D Scholar
>> >>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar
>> >>>>>> University
>> >>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>> >>>>>> Ph. No: +91-9897065948
>> >>>>>>
>> >>>>>> ............................................................
>> >>>>> ...............................................................
>> >>>>>> *A man is the best judge of himself and he has to pay the price for
>> >>>>>> what
>> >>>>> he
>> >>>>>> does.*
>> >>>>>> ............................................................
>> >>>>> ...............................................................
>> >>>>>>
>> >>>>>>
>> >>>>>>
>> >>>>>> _______________________________________________
>> >>>>>> Wekalist mailing list
>> >>>>>> Send posts to: [hidden email]
>> >>>>>> List info and subscription status: https://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: [hidden email]
>> >>>>> List info and subscription status: https://list.waikato.ac.nz/
>> >>>>> mailman/listinfo/wekalist
>> >>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>> >>>>> ml/weka/mailinglist_etiquette.html
>> >>>>>
>> >>>>
>> >>>
>> >>>
>> >>> --
>> >>> Thanks & Regards
>> >>> Bhupesh Rawat.
>> >>> Ph.D Scholar
>> >>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>> >>> Vidya Vihar,Rai Bareilly road(Lucknow)
>> >>> Ph. No: +91-9897065948
>> >>>
>> >>> ...........................................................................................................................
>> >>> *A man is the best judge of himself and he has to pay the price for
>> >>> what
>> >>> he
>> >>> does.*
>> >>> ...........................................................................................................................
>> >>> <students' data after
>> >>> preprocessin.xlsx>_______________________________________________
>> >>> Wekalist mailing list
>> >>> Send posts to: [hidden email]
>> >>> List info and subscription status:
>> >>> https://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: [hidden email]
>> >> List info and subscription status:
>> >> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>> >> List etiquette:
>> >> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>> >>
>> >
>> >
>> > --
>> > Thanks & Regards
>> > Bhupesh Rawat.
>> > Ph.D Scholar
>> > Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>> > Vidya Vihar,Rai Bareilly road(Lucknow)
>> > Ph. No: +91-9897065948
>> >
>> > ...........................................................................................................................
>> > *A man is the best judge of himself and he has to pay the price for what
>> > he
>> > does.*
>> > ...........................................................................................................................
>> > _______________________________________________
>> > Wekalist mailing list
>> > Send posts to: [hidden email]
>> > List info and subscription status:
>> > https://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: [hidden email]
>> List info and subscription status:
>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>> List etiquette:
>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>
>>
>>
>> --
>> Thanks & Regards
>> Bhupesh Rawat.
>> Ph.D Scholar
>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>> Vidya Vihar,Rai Bareilly road(Lucknow)
>> Ph. No: +91-9897065948
>>
>> ...........................................................................................................................
>> *A man is the best judge of himself and he has to pay the price for what
>> he
>> does.*
>> ...........................................................................................................................
>>
>>
>>
>> <knowledge flow
>> interuppted.docx>_______________________________________________
>> Wekalist mailing list
>> Send posts to: [hidden email]
>> List info and subscription status:
>> https://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: [hidden email]
> List info and subscription status:
> https://list.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>


--
Thanks & Regards
Bhupesh Rawat.
Ph.D Scholar
Department of Computer Science,Babasaheb Bhimrao Ambedkar University
Vidya Vihar,Rai Bareilly road(Lucknow)
Ph. No: +91-9897065948

...........................................................................................................................
*A man is the best judge of himself and he has to pay the price for what he
does.*
...........................................................................................................................
_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Eibe Frank-2
Administrator
In WEKA 3.8/3.9, under

  filters.unsupervised.attribute.CartesianProduct

Cheers,
Eibe

> On 1/03/2017, at 6:13 PM, Bhupesh Rawat <[hidden email]> wrote:
>
> Thank you Sir, the problem has been fixed.
>
> Moreover i would also like to use the combination of attributes for
> which  you suggested  the CartesionProduct filter. Where could i find
> this option?
>
> On 2/28/17, Eibe Frank <[hidden email]> wrote:
>> What does the log say (see the “log” tab next to the “status” tab)?
>>
>> Cheers,
>> Eibe
>>
>>> On 27/02/2017, at 11:56 PM, Bhupesh Rawat <[hidden email]> wrote:
>>>
>>> Sir,
>>> When I use the KnowledgeFlow GUI the status shown by two of the components
>>> is interrupted(namely crossvalidationfoldmaker and J48) as shown in the
>>> attached file. How to fix it?
>>>
>>> On Mon, Feb 27, 2017 at 3:08 AM, Eibe Frank <[hidden email]> wrote:
>>> In the Explorer, there is no way to turn off evaluation completely. You
>>> could use the command-line interface or the KnowledgeFlow GUI though.
>>>
>>> Having said this, if you evaluate on the training set, the runtime
>>> overhead is quite small if you apply a rule learner.
>>>
>>> Note also that the Explorer always outputs the classification model for
>>> the *full* dataset loaded into the Preprocess panel, regardless of which
>>> evaluation metric you choose, i.e., you will get the rule set for the full
>>> dataset regardless of the evaluation method you use.
>>>
>>> Cheers,
>>> Eibe
>>>
>>>> On 26/02/2017, at 8:07 PM, Bhupesh Rawat <[hidden email]> wrote:
>>>>
>>>> Sir,
>>>>
>>>> How could i perform these two task seperately(applying classification
>>>> rule learner and estimating classification accuracy). The accuracy is
>>>> estimated each time i run the classifier on the dataset.
>>>>
>>>>
>>>>
>>>> On 2/24/17, Eibe Frank <[hidden email]> wrote:
>>>>> No, not really. However, the dataset is quite small. You could just run
>>>>> a
>>>>> classification rule learner such as JRip or PART on the data, treating
>>>>> each
>>>>> of the attributes in turn as the class attribute. Then you can estimate
>>>>> classification accuracy using cross-validation.
>>>>>
>>>>> You could also create combinations of attributes using the
>>>>> CartesionProduct
>>>>> filter.
>>>>>
>>>>> Cheers,
>>>>> Eibe
>>>>>
>>>>>> On 24/02/2017, at 3:11 AM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>
>>>>>> I have a small dataset which contains student enrolment data in
>>>>>> various courses. If a student has selected a particular course it is
>>>>>> indicated by ‘Y’ else ‘N’ is used. I have also attached a file for
>>>>>> better understanding of the dataset. I am interested in knowing if it
>>>>>> is possible to measure the accuracy of the association rules with this
>>>>>> dataset by the proposed approach in your paper.
>>>>>>
>>>>>> On 2/23/17, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>> Thank you so much for the response!!
>>>>>>> On Feb 23, 2017 8:26 AM, "Eibe Frank" <[hidden email]> wrote:
>>>>>>>
>>>>>>>> You mean beyond confidence, lift, or one the other metrics that you
>>>>>>>> can
>>>>>>>> get in the output of each rule? This is a tough question. One way
>>>>>>>> may be
>>>>>>>> to
>>>>>>>> use the association rule mining algorithm to build classification
>>>>>>>> rules
>>>>>>>> and
>>>>>>>> then evaluate the accuracy of those classification rules. We had a
>>>>>>>> paper
>>>>>>>> on
>>>>>>>> this quite a while back:
>>>>>>>>
>>>>>>>> Mutter, S., Hall, M., & Frank, E. (2004, December). Using
>>>>>>>> classification
>>>>>>>> to evaluate the output of confidence-based association rule mining.
>>>>>>>> In
>>>>>>>> Australasian Joint Conference on Artificial Intelligence (pp.
>>>>>>>> 538-549).
>>>>>>>> Springer Berlin Heidelberg.
>>>>>>>>
>>>>>>>> I suppose you could also evaluate the individual association rules
>>>>>>>> on a
>>>>>>>> separate test set, by computing the confidence measure, etc., on the
>>>>>>>> test
>>>>>>>> set for each rule, but this functionality is not provided by WEKA.
>>>>>>>>
>>>>>>>> Cheers,
>>>>>>>> Eibe
>>>>>>>>
>>>>>>>>> On 23/02/2017, at 12:46 AM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>>>>
>>>>>>>>> Dear Sir/Madam
>>>>>>>>>
>>>>>>>>> I have discovered some rules through weka. Could you tell me how to
>>>>>>>> measure  the accuracy of those rules.
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> On Tue, Feb 14, 2017 at 3:44 AM, Peter Reutemann
>>>>>>>>> <[hidden email]>
>>>>>>>> wrote:
>>>>>>>>>> The size of the final Apriori model as a serialised Java object
>>>>>>>>>> can
>>>>>>>>>> be
>>>>>>>> established saving it to a file and considering the file size. Note
>>>>>>>> that
>>>>>>>> this is different from the size of the object in memory (see, e.g.,
>>>>>>>> http://stackoverflow.com/questions/7146559/serialized-
>>>>>>>> object-size-vs-in-memory-object-size-in-java#7146941).
>>>>>>>>>>
>>>>>>>>>> I don’t know of a good way to measure peak memory consumption of a
>>>>>>>> Java program (after garbage collection). A crude way would be to run
>>>>>>>> the
>>>>>>>> program from the command-line (to avoid overhead associated with the
>>>>>>>> GUIs)
>>>>>>>> with different maximum heap sizes, e.g., increasing the heap size
>>>>>>>> until
>>>>>>>> the
>>>>>>>> program runs through. Another option is to look at the heap size in
>>>>>>>> a
>>>>>>>> profiler (e.g., visualvm), enforcing garbage collection before a
>>>>>>>> readout.
>>>>>>>>>
>>>>>>>>> You can use the sizeofag javaagent for determining the size of a
>>>>>>>>> Java
>>>>>>>> object:
>>>>>>>>> https://github.com/fracpete/sizeofag
>>>>>>>>>
>>>>>>>>> Credits to Maxim Zakharenkov, who wrote the original code.
>>>>>>>>>
>>>>>>>>> Cheers, Peter
>>>>>>>>> --
>>>>>>>>> Peter Reutemann
>>>>>>>>> Dept. of Computer Science
>>>>>>>>> University of Waikato, NZ
>>>>>>>>> +64 (7) 858-5174
>>>>>>>>> http://www.cms.waikato.ac.nz/~fracpete/
>>>>>>>>> http://www.data-mining.co.nz/
>>>>>>>>> _______________________________________________
>>>>>>>>> Wekalist mailing list
>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>>>>>> mailman/listinfo/wekalist
>>>>>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>>>>>> ml/weka/mailinglist_etiquette.html
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> --
>>>>>>>>> Thanks & Regards
>>>>>>>>> Bhupesh Rawat.
>>>>>>>>> Ph.D Scholar
>>>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar
>>>>>>>>> University
>>>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>>>> Ph. No: +91-9897065948
>>>>>>>>>
>>>>>>>>> ............................................................
>>>>>>>> ...............................................................
>>>>>>>>> *A man is the best judge of himself and he has to pay the price for
>>>>>>>>> what
>>>>>>>> he
>>>>>>>>> does.*
>>>>>>>>> ............................................................
>>>>>>>> ...............................................................
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> _______________________________________________
>>>>>>>>> Wekalist mailing list
>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>> List info and subscription status: https://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: [hidden email]
>>>>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>>>>>> mailman/listinfo/wekalist
>>>>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>>>>>> ml/weka/mailinglist_etiquette.html
>>>>>>>>
>>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> Thanks & Regards
>>>>>> Bhupesh Rawat.
>>>>>> Ph.D Scholar
>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>> Ph. No: +91-9897065948
>>>>>>
>>>>>> ...........................................................................................................................
>>>>>> *A man is the best judge of himself and he has to pay the price for
>>>>>> what
>>>>>> he
>>>>>> does.*
>>>>>> ...........................................................................................................................
>>>>>> <students' data after
>>>>>> preprocessin.xlsx>_______________________________________________
>>>>>> Wekalist mailing list
>>>>>> Send posts to: [hidden email]
>>>>>> List info and subscription status:
>>>>>> https://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: [hidden email]
>>>>> List info and subscription status:
>>>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>>>> List etiquette:
>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>
>>>>
>>>>
>>>> --
>>>> Thanks & Regards
>>>> Bhupesh Rawat.
>>>> Ph.D Scholar
>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>> Ph. No: +91-9897065948
>>>>
>>>> ...........................................................................................................................
>>>> *A man is the best judge of himself and he has to pay the price for what
>>>> he
>>>> does.*
>>>> ...........................................................................................................................
>>>> _______________________________________________
>>>> Wekalist mailing list
>>>> Send posts to: [hidden email]
>>>> List info and subscription status:
>>>> https://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: [hidden email]
>>> List info and subscription status:
>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>> List etiquette:
>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>
>>>
>>>
>>> --
>>> Thanks & Regards
>>> Bhupesh Rawat.
>>> Ph.D Scholar
>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>> Ph. No: +91-9897065948
>>>
>>> ...........................................................................................................................
>>> *A man is the best judge of himself and he has to pay the price for what
>>> he
>>> does.*
>>> ...........................................................................................................................
>>>
>>>
>>>
>>> <knowledge flow
>>> interuppted.docx>_______________________________________________
>>> Wekalist mailing list
>>> Send posts to: [hidden email]
>>> List info and subscription status:
>>> https://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: [hidden email]
>> List info and subscription status:
>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>> List etiquette:
>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>
>
>
> --
> Thanks & Regards
> Bhupesh Rawat.
> Ph.D Scholar
> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
> Vidya Vihar,Rai Bareilly road(Lucknow)
> Ph. No: +91-9897065948
>
> ...........................................................................................................................
> *A man is the best judge of himself and he has to pay the price for what he
> does.*
> ...........................................................................................................................
> _______________________________________________
> Wekalist mailing list
> Send posts to: [hidden email]
> List info and subscription status: https://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: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Bhupesh Rawat
Sir,

How to choose combination of attribute as a class attribute with Jrip
or PART in weka 3.8.

Moreover i tried Predictive apriori on the dataset and as a result i
found some rules with their respective accuracy. How reliable are
those rules based on this accuracy.





On Thu, Mar 2, 2017 at 2:10 AM, Eibe Frank <[hidden email]> wrote:

> In WEKA 3.8/3.9, under
>
>   filters.unsupervised.attribute.CartesianProduct
>
> Cheers,
> Eibe
>
> > On 1/03/2017, at 6:13 PM, Bhupesh Rawat <[hidden email]> wrote:
> >
> > Thank you Sir, the problem has been fixed.
> >
> > Moreover i would also like to use the combination of attributes for
> > which  you suggested  the CartesionProduct filter. Where could i find
> > this option?
> >
> > On 2/28/17, Eibe Frank <[hidden email]> wrote:
> >> What does the log say (see the “log” tab next to the “status” tab)?
> >>
> >> Cheers,
> >> Eibe
> >>
> >>> On 27/02/2017, at 11:56 PM, Bhupesh Rawat <[hidden email]> wrote:
> >>>
> >>> Sir,
> >>> When I use the KnowledgeFlow GUI the status shown by two of the
> components
> >>> is interrupted(namely crossvalidationfoldmaker and J48) as shown in the
> >>> attached file. How to fix it?
> >>>
> >>> On Mon, Feb 27, 2017 at 3:08 AM, Eibe Frank <[hidden email]>
> wrote:
> >>> In the Explorer, there is no way to turn off evaluation completely. You
> >>> could use the command-line interface or the KnowledgeFlow GUI though.
> >>>
> >>> Having said this, if you evaluate on the training set, the runtime
> >>> overhead is quite small if you apply a rule learner.
> >>>
> >>> Note also that the Explorer always outputs the classification model for
> >>> the *full* dataset loaded into the Preprocess panel, regardless of
> which
> >>> evaluation metric you choose, i.e., you will get the rule set for the
> full
> >>> dataset regardless of the evaluation method you use.
> >>>
> >>> Cheers,
> >>> Eibe
> >>>
> >>>> On 26/02/2017, at 8:07 PM, Bhupesh Rawat <[hidden email]> wrote:
> >>>>
> >>>> Sir,
> >>>>
> >>>> How could i perform these two task seperately(applying classification
> >>>> rule learner and estimating classification accuracy). The accuracy is
> >>>> estimated each time i run the classifier on the dataset.
> >>>>
> >>>>
> >>>>
> >>>> On 2/24/17, Eibe Frank <[hidden email]> wrote:
> >>>>> No, not really. However, the dataset is quite small. You could just
> run
> >>>>> a
> >>>>> classification rule learner such as JRip or PART on the data,
> treating
> >>>>> each
> >>>>> of the attributes in turn as the class attribute. Then you can
> estimate
> >>>>> classification accuracy using cross-validation.
> >>>>>
> >>>>> You could also create combinations of attributes using the
> >>>>> CartesionProduct
> >>>>> filter.
> >>>>>
> >>>>> Cheers,
> >>>>> Eibe
> >>>>>
> >>>>>> On 24/02/2017, at 3:11 AM, Bhupesh Rawat <[hidden email]> wrote:
> >>>>>>
> >>>>>> I have a small dataset which contains student enrolment data in
> >>>>>> various courses. If a student has selected a particular course it is
> >>>>>> indicated by ‘Y’ else ‘N’ is used. I have also attached a file for
> >>>>>> better understanding of the dataset. I am interested in knowing if
> it
> >>>>>> is possible to measure the accuracy of the association rules with
> this
> >>>>>> dataset by the proposed approach in your paper.
> >>>>>>
> >>>>>> On 2/23/17, Bhupesh Rawat <[hidden email]> wrote:
> >>>>>>> Thank you so much for the response!!
> >>>>>>> On Feb 23, 2017 8:26 AM, "Eibe Frank" <[hidden email]> wrote:
> >>>>>>>
> >>>>>>>> You mean beyond confidence, lift, or one the other metrics that
> you
> >>>>>>>> can
> >>>>>>>> get in the output of each rule? This is a tough question. One way
> >>>>>>>> may be
> >>>>>>>> to
> >>>>>>>> use the association rule mining algorithm to build classification
> >>>>>>>> rules
> >>>>>>>> and
> >>>>>>>> then evaluate the accuracy of those classification rules. We had a
> >>>>>>>> paper
> >>>>>>>> on
> >>>>>>>> this quite a while back:
> >>>>>>>>
> >>>>>>>> Mutter, S., Hall, M., & Frank, E. (2004, December). Using
> >>>>>>>> classification
> >>>>>>>> to evaluate the output of confidence-based association rule
> mining.
> >>>>>>>> In
> >>>>>>>> Australasian Joint Conference on Artificial Intelligence (pp.
> >>>>>>>> 538-549).
> >>>>>>>> Springer Berlin Heidelberg.
> >>>>>>>>
> >>>>>>>> I suppose you could also evaluate the individual association rules
> >>>>>>>> on a
> >>>>>>>> separate test set, by computing the confidence measure, etc., on
> the
> >>>>>>>> test
> >>>>>>>> set for each rule, but this functionality is not provided by WEKA.
> >>>>>>>>
> >>>>>>>> Cheers,
> >>>>>>>> Eibe
> >>>>>>>>
> >>>>>>>>> On 23/02/2017, at 12:46 AM, Bhupesh Rawat <[hidden email]>
> wrote:
> >>>>>>>>>
> >>>>>>>>> Dear Sir/Madam
> >>>>>>>>>
> >>>>>>>>> I have discovered some rules through weka. Could you tell me how
> to
> >>>>>>>> measure  the accuracy of those rules.
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>> On Tue, Feb 14, 2017 at 3:44 AM, Peter Reutemann
> >>>>>>>>> <[hidden email]>
> >>>>>>>> wrote:
> >>>>>>>>>> The size of the final Apriori model as a serialised Java object
> >>>>>>>>>> can
> >>>>>>>>>> be
> >>>>>>>> established saving it to a file and considering the file size.
> Note
> >>>>>>>> that
> >>>>>>>> this is different from the size of the object in memory (see,
> e.g.,
> >>>>>>>> http://stackoverflow.com/questions/7146559/serialized-
> >>>>>>>> object-size-vs-in-memory-object-size-in-java#7146941).
> >>>>>>>>>>
> >>>>>>>>>> I don’t know of a good way to measure peak memory consumption
> of a
> >>>>>>>> Java program (after garbage collection). A crude way would be to
> run
> >>>>>>>> the
> >>>>>>>> program from the command-line (to avoid overhead associated with
> the
> >>>>>>>> GUIs)
> >>>>>>>> with different maximum heap sizes, e.g., increasing the heap size
> >>>>>>>> until
> >>>>>>>> the
> >>>>>>>> program runs through. Another option is to look at the heap size
> in
> >>>>>>>> a
> >>>>>>>> profiler (e.g., visualvm), enforcing garbage collection before a
> >>>>>>>> readout.
> >>>>>>>>>
> >>>>>>>>> You can use the sizeofag javaagent for determining the size of a
> >>>>>>>>> Java
> >>>>>>>> object:
> >>>>>>>>> https://github.com/fracpete/sizeofag
> >>>>>>>>>
> >>>>>>>>> Credits to Maxim Zakharenkov, who wrote the original code.
> >>>>>>>>>
> >>>>>>>>> Cheers, Peter
> >>>>>>>>> --
> >>>>>>>>> Peter Reutemann
> >>>>>>>>> Dept. of Computer Science
> >>>>>>>>> University of Waikato, NZ
> >>>>>>>>> +64 (7) 858-5174
> >>>>>>>>> http://www.cms.waikato.ac.nz/~fracpete/
> >>>>>>>>> http://www.data-mining.co.nz/
> >>>>>>>>> _______________________________________________
> >>>>>>>>> Wekalist mailing list
> >>>>>>>>> Send posts to: [hidden email]
> >>>>>>>>> List info and subscription status: https://list.waikato.ac.nz/
> >>>>>>>> mailman/listinfo/wekalist
> >>>>>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
> >>>>>>>> ml/weka/mailinglist_etiquette.html
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>> --
> >>>>>>>>> Thanks & Regards
> >>>>>>>>> Bhupesh Rawat.
> >>>>>>>>> Ph.D Scholar
> >>>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar
> >>>>>>>>> University
> >>>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
> >>>>>>>>> Ph. No: +91-9897065948
> >>>>>>>>>
> >>>>>>>>> ............................................................
> >>>>>>>> ...............................................................
> >>>>>>>>> *A man is the best judge of himself and he has to pay the price
> for
> >>>>>>>>> what
> >>>>>>>> he
> >>>>>>>>> does.*
> >>>>>>>>> ............................................................
> >>>>>>>> ...............................................................
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>> _______________________________________________
> >>>>>>>>> Wekalist mailing list
> >>>>>>>>> Send posts to: [hidden email]
> >>>>>>>>> List info and subscription status: https://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: [hidden email]
> >>>>>>>> List info and subscription status: https://list.waikato.ac.nz/
> >>>>>>>> mailman/listinfo/wekalist
> >>>>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
> >>>>>>>> ml/weka/mailinglist_etiquette.html
> >>>>>>>>
> >>>>>>>
> >>>>>>
> >>>>>>
> >>>>>> --
> >>>>>> Thanks & Regards
> >>>>>> Bhupesh Rawat.
> >>>>>> Ph.D Scholar
> >>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
> >>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
> >>>>>> Ph. No: +91-9897065948
> >>>>>>
> >>>>>> ............................................................
> ...............................................................
> >>>>>> *A man is the best judge of himself and he has to pay the price for
> >>>>>> what
> >>>>>> he
> >>>>>> does.*
> >>>>>> ............................................................
> ...............................................................
> >>>>>> <students' data after
> >>>>>> preprocessin.xlsx>_______________________________________________
> >>>>>> Wekalist mailing list
> >>>>>> Send posts to: [hidden email]
> >>>>>> List info and subscription status:
> >>>>>> https://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: [hidden email]
> >>>>> List info and subscription status:
> >>>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
> >>>>> List etiquette:
> >>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> >>>>>
> >>>>
> >>>>
> >>>> --
> >>>> Thanks & Regards
> >>>> Bhupesh Rawat.
> >>>> Ph.D Scholar
> >>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
> >>>> Vidya Vihar,Rai Bareilly road(Lucknow)
> >>>> Ph. No: +91-9897065948
> >>>>
> >>>> ............................................................
> ...............................................................
> >>>> *A man is the best judge of himself and he has to pay the price for
> what
> >>>> he
> >>>> does.*
> >>>> ............................................................
> ...............................................................
> >>>> _______________________________________________
> >>>> Wekalist mailing list
> >>>> Send posts to: [hidden email]
> >>>> List info and subscription status:
> >>>> https://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: [hidden email]
> >>> List info and subscription status:
> >>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
> >>> List etiquette:
> >>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> >>>
> >>>
> >>>
> >>> --
> >>> Thanks & Regards
> >>> Bhupesh Rawat.
> >>> Ph.D Scholar
> >>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
> >>> Vidya Vihar,Rai Bareilly road(Lucknow)
> >>> Ph. No: +91-9897065948
> >>>
> >>> ............................................................
> ...............................................................
> >>> *A man is the best judge of himself and he has to pay the price for
> what
> >>> he
> >>> does.*
> >>> ............................................................
> ...............................................................
> >>>
> >>>
> >>>
> >>> <knowledge flow
> >>> interuppted.docx>_______________________________________________
> >>> Wekalist mailing list
> >>> Send posts to: [hidden email]
> >>> List info and subscription status:
> >>> https://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: [hidden email]
> >> List info and subscription status:
> >> https://list.waikato.ac.nz/mailman/listinfo/wekalist
> >> List etiquette:
> >> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
> >>
> >
> >
> > --
> > Thanks & Regards
> > Bhupesh Rawat.
> > Ph.D Scholar
> > Department of Computer Science,Babasaheb Bhimrao Ambedkar University
> > Vidya Vihar,Rai Bareilly road(Lucknow)
> > Ph. No: +91-9897065948
> >
> > ............................................................
> ...............................................................
> > *A man is the best judge of himself and he has to pay the price for what
> he
> > does.*
> > ............................................................
> ...............................................................
> > _______________________________________________
> > Wekalist mailing list
> > Send posts to: [hidden email]
> > List info and subscription status: https://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: [hidden email]
> List info and subscription status: https://list.waikato.ac.nz/
> mailman/listinfo/wekalist
> List etiquette: http://www.cs.waikato.ac.nz/~
> ml/weka/mailinglist_etiquette.html
>



--
Thanks & Regards
Bhupesh Rawat.
Ph.D Scholar
Department of Computer Science,Babasaheb Bhimrao Ambedkar University
Vidya Vihar,Rai Bareilly road(Lucknow)
Ph. No: +91-9897065948

...........................................................................................................................
*A man is the best judge of himself and he has to pay the price for what he
does.*
...........................................................................................................................
_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Eibe Frank-2
Administrator
You can create a new attribute by combining nominal attributes using the CartesianProduct filter.

Regarding the reliability of the rules, take a look at the literature for "predictive apriori" on Google Scholar. I don't know if there have been any extensive studies.

To get a rough idea of how well PredictiveApriori works for your data, regardless of the accuracy of individual rules considered in isolation, you could apply it to mine class association rules with the JCBA classifier (from the classAssociationRules package) and use cross-validation for evaluation, similar to what we did in our paper. Obviously, you will have to create an appropriate class attribute for each attribute/attribute combination that you are interested in (possibly using CartesianProduct).

Here is an example command-line, running JCBA with PredictiveApriori on the vote data (using the default class attribute). I got it to only output the top two rules for simplicity:

===================

java weka.Run .JCBA -A ".PredictiveApriori -N 2" -t ~/datasets/UCI/vote.arff

Options: -A ".PredictiveApriori -N 2"


Classification Rules (ordered):
==========================

1. physician-fee-freeze=n 3 0 adoption-of-the-budget-resolution=y 2 1  ==> Class=democrat     acc:(0.99),  (219),  
2. crime=n 13 0 el-salvador-aid=n 4 0 adoption-of-the-budget-resolution=y 2 1  ==> Class=democrat     acc:(0.99),  (144),  


Default Class: Class=republican

Additional Information:
Number of Classification Associations Rules generated by Rule Miner: 2
Number of Classification Rules: 2

Mining Time in sec.: 7.867
Pruning Time in sec. : 0.033


Time taken to build model: 7.91 seconds
Time taken to test model on training data: 0.02 seconds

=== Error on training data ===

Correctly Classified Instances         389               89.4253 %
Incorrectly Classified Instances        46               10.5747 %
Kappa statistic                          0.7877
Mean absolute error                      0.1057
Root mean squared error                  0.3252
Relative absolute error                 22.2991 %
Root relative squared error             66.7902 %
Total Number of Instances              435    


=== Detailed Accuracy By Class ===

                 TP Rate  FP Rate  Precision  Recall   F-Measure  MCC      ROC Area  PRC Area  Class
                 0.828    0.000    1.000      0.828    0.906      0.806    0.914     0.933     democrat
                 1.000    0.172    0.785      1.000    0.880      0.806    0.914     0.785     republican
Weighted Avg.    0.894    0.067    0.917      0.894    0.896      0.806    0.914     0.876    


=== Confusion Matrix ===

   a   b   <-- classified as
 221  46 |   a = democrat
   0 168 |   b = republican



=== Stratified cross-validation ===

Correctly Classified Instances         391               89.8851 %
Incorrectly Classified Instances        44               10.1149 %
Kappa statistic                          0.7957
Mean absolute error                      0.1011
Root mean squared error                  0.318
Relative absolute error                 21.3284 %
Root relative squared error             65.3201 %
Total Number of Instances              435    


=== Detailed Accuracy By Class ===

                 TP Rate  FP Rate  Precision  Recall   F-Measure  MCC      ROC Area  PRC Area  Class
                 0.843    0.012    0.991      0.843    0.911      0.810    0.915     0.932     democrat
                 0.988    0.157    0.798      0.988    0.883      0.810    0.915     0.793     republican
Weighted Avg.    0.899    0.068    0.917      0.899    0.900      0.810    0.915     0.878    


=== Confusion Matrix ===

   a   b   <-- classified as
 225  42 |   a = democrat
   2 166 |   b = republican

===================

The observed precision of classifications for class democrat estimated by cross-validation (under "Detailed Accuracy By Class") is quite close to the accuracy estimates listed for the individual rules in the initial output of the class association rules (0.99), so we can be reasonably confident in this case that the rules are very accurate.

This process doesn't give you independent accuracy estimates for individual rules though. Assuming you have a reasonably large test set, you could code up individual rules in PMML and use the PMML classifier in WEKA for each rule to evaluate it on the test set (http://wiki.pentaho.com/display/DATAMINING/PMML+Support+in+Weka). However, you mentioned that you have a small dataset so this is probably not an option for you.

Cheers,
Eibe

> On 5 Mar 2017, at 03:39, Bhupesh Rawat <[hidden email]> wrote:
>
> Sir,
>
> How to choose combination of attribute as a class attribute with Jrip
> or PART in weka 3.8.
>
> Moreover i tried Predictive apriori on the dataset and as a result i
> found some rules with their respective accuracy. How reliable are
> those rules based on this accuracy.
>
>
>
>
>
> On Thu, Mar 2, 2017 at 2:10 AM, Eibe Frank <[hidden email]> wrote:
>
>> In WEKA 3.8/3.9, under
>>
>>  filters.unsupervised.attribute.CartesianProduct
>>
>> Cheers,
>> Eibe
>>
>>> On 1/03/2017, at 6:13 PM, Bhupesh Rawat <[hidden email]> wrote:
>>>
>>> Thank you Sir, the problem has been fixed.
>>>
>>> Moreover i would also like to use the combination of attributes for
>>> which  you suggested  the CartesionProduct filter. Where could i find
>>> this option?
>>>
>>> On 2/28/17, Eibe Frank <[hidden email]> wrote:
>>>> What does the log say (see the “log” tab next to the “status” tab)?
>>>>
>>>> Cheers,
>>>> Eibe
>>>>
>>>>> On 27/02/2017, at 11:56 PM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>
>>>>> Sir,
>>>>> When I use the KnowledgeFlow GUI the status shown by two of the
>> components
>>>>> is interrupted(namely crossvalidationfoldmaker and J48) as shown in the
>>>>> attached file. How to fix it?
>>>>>
>>>>> On Mon, Feb 27, 2017 at 3:08 AM, Eibe Frank <[hidden email]>
>> wrote:
>>>>> In the Explorer, there is no way to turn off evaluation completely. You
>>>>> could use the command-line interface or the KnowledgeFlow GUI though.
>>>>>
>>>>> Having said this, if you evaluate on the training set, the runtime
>>>>> overhead is quite small if you apply a rule learner.
>>>>>
>>>>> Note also that the Explorer always outputs the classification model for
>>>>> the *full* dataset loaded into the Preprocess panel, regardless of
>> which
>>>>> evaluation metric you choose, i.e., you will get the rule set for the
>> full
>>>>> dataset regardless of the evaluation method you use.
>>>>>
>>>>> Cheers,
>>>>> Eibe
>>>>>
>>>>>> On 26/02/2017, at 8:07 PM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>
>>>>>> Sir,
>>>>>>
>>>>>> How could i perform these two task seperately(applying classification
>>>>>> rule learner and estimating classification accuracy). The accuracy is
>>>>>> estimated each time i run the classifier on the dataset.
>>>>>>
>>>>>>
>>>>>>
>>>>>> On 2/24/17, Eibe Frank <[hidden email]> wrote:
>>>>>>> No, not really. However, the dataset is quite small. You could just
>> run
>>>>>>> a
>>>>>>> classification rule learner such as JRip or PART on the data,
>> treating
>>>>>>> each
>>>>>>> of the attributes in turn as the class attribute. Then you can
>> estimate
>>>>>>> classification accuracy using cross-validation.
>>>>>>>
>>>>>>> You could also create combinations of attributes using the
>>>>>>> CartesionProduct
>>>>>>> filter.
>>>>>>>
>>>>>>> Cheers,
>>>>>>> Eibe
>>>>>>>
>>>>>>>> On 24/02/2017, at 3:11 AM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>>>
>>>>>>>> I have a small dataset which contains student enrolment data in
>>>>>>>> various courses. If a student has selected a particular course it is
>>>>>>>> indicated by ‘Y’ else ‘N’ is used. I have also attached a file for
>>>>>>>> better understanding of the dataset. I am interested in knowing if
>> it
>>>>>>>> is possible to measure the accuracy of the association rules with
>> this
>>>>>>>> dataset by the proposed approach in your paper.
>>>>>>>>
>>>>>>>> On 2/23/17, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>>>> Thank you so much for the response!!
>>>>>>>>> On Feb 23, 2017 8:26 AM, "Eibe Frank" <[hidden email]> wrote:
>>>>>>>>>
>>>>>>>>>> You mean beyond confidence, lift, or one the other metrics that
>> you
>>>>>>>>>> can
>>>>>>>>>> get in the output of each rule? This is a tough question. One way
>>>>>>>>>> may be
>>>>>>>>>> to
>>>>>>>>>> use the association rule mining algorithm to build classification
>>>>>>>>>> rules
>>>>>>>>>> and
>>>>>>>>>> then evaluate the accuracy of those classification rules. We had a
>>>>>>>>>> paper
>>>>>>>>>> on
>>>>>>>>>> this quite a while back:
>>>>>>>>>>
>>>>>>>>>> Mutter, S., Hall, M., & Frank, E. (2004, December). Using
>>>>>>>>>> classification
>>>>>>>>>> to evaluate the output of confidence-based association rule
>> mining.
>>>>>>>>>> In
>>>>>>>>>> Australasian Joint Conference on Artificial Intelligence (pp.
>>>>>>>>>> 538-549).
>>>>>>>>>> Springer Berlin Heidelberg.
>>>>>>>>>>
>>>>>>>>>> I suppose you could also evaluate the individual association rules
>>>>>>>>>> on a
>>>>>>>>>> separate test set, by computing the confidence measure, etc., on
>> the
>>>>>>>>>> test
>>>>>>>>>> set for each rule, but this functionality is not provided by WEKA.
>>>>>>>>>>
>>>>>>>>>> Cheers,
>>>>>>>>>> Eibe
>>>>>>>>>>
>>>>>>>>>>> On 23/02/2017, at 12:46 AM, Bhupesh Rawat <[hidden email]>
>> wrote:
>>>>>>>>>>>
>>>>>>>>>>> Dear Sir/Madam
>>>>>>>>>>>
>>>>>>>>>>> I have discovered some rules through weka. Could you tell me how
>> to
>>>>>>>>>> measure  the accuracy of those rules.
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> On Tue, Feb 14, 2017 at 3:44 AM, Peter Reutemann
>>>>>>>>>>> <[hidden email]>
>>>>>>>>>> wrote:
>>>>>>>>>>>> The size of the final Apriori model as a serialised Java object
>>>>>>>>>>>> can
>>>>>>>>>>>> be
>>>>>>>>>> established saving it to a file and considering the file size.
>> Note
>>>>>>>>>> that
>>>>>>>>>> this is different from the size of the object in memory (see,
>> e.g.,
>>>>>>>>>> http://stackoverflow.com/questions/7146559/serialized-
>>>>>>>>>> object-size-vs-in-memory-object-size-in-java#7146941).
>>>>>>>>>>>>
>>>>>>>>>>>> I don’t know of a good way to measure peak memory consumption
>> of a
>>>>>>>>>> Java program (after garbage collection). A crude way would be to
>> run
>>>>>>>>>> the
>>>>>>>>>> program from the command-line (to avoid overhead associated with
>> the
>>>>>>>>>> GUIs)
>>>>>>>>>> with different maximum heap sizes, e.g., increasing the heap size
>>>>>>>>>> until
>>>>>>>>>> the
>>>>>>>>>> program runs through. Another option is to look at the heap size
>> in
>>>>>>>>>> a
>>>>>>>>>> profiler (e.g., visualvm), enforcing garbage collection before a
>>>>>>>>>> readout.
>>>>>>>>>>>
>>>>>>>>>>> You can use the sizeofag javaagent for determining the size of a
>>>>>>>>>>> Java
>>>>>>>>>> object:
>>>>>>>>>>> https://github.com/fracpete/sizeofag
>>>>>>>>>>>
>>>>>>>>>>> Credits to Maxim Zakharenkov, who wrote the original code.
>>>>>>>>>>>
>>>>>>>>>>> Cheers, Peter
>>>>>>>>>>> --
>>>>>>>>>>> Peter Reutemann
>>>>>>>>>>> Dept. of Computer Science
>>>>>>>>>>> University of Waikato, NZ
>>>>>>>>>>> +64 (7) 858-5174
>>>>>>>>>>> http://www.cms.waikato.ac.nz/~fracpete/
>>>>>>>>>>> http://www.data-mining.co.nz/
>>>>>>>>>>> _______________________________________________
>>>>>>>>>>> Wekalist mailing list
>>>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>>>>>>>> mailman/listinfo/wekalist
>>>>>>>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>>>>>>>> ml/weka/mailinglist_etiquette.html
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> --
>>>>>>>>>>> Thanks & Regards
>>>>>>>>>>> Bhupesh Rawat.
>>>>>>>>>>> Ph.D Scholar
>>>>>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar
>>>>>>>>>>> University
>>>>>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>>>>>> Ph. No: +91-9897065948
>>>>>>>>>>>
>>>>>>>>>>> ............................................................
>>>>>>>>>> ...............................................................
>>>>>>>>>>> *A man is the best judge of himself and he has to pay the price
>> for
>>>>>>>>>>> what
>>>>>>>>>> he
>>>>>>>>>>> does.*
>>>>>>>>>>> ............................................................
>>>>>>>>>> ...............................................................
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> _______________________________________________
>>>>>>>>>>> Wekalist mailing list
>>>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>>>> List info and subscription status: https://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: [hidden email]
>>>>>>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>>>>>>>> mailman/listinfo/wekalist
>>>>>>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>>>>>>>> ml/weka/mailinglist_etiquette.html
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> --
>>>>>>>> Thanks & Regards
>>>>>>>> Bhupesh Rawat.
>>>>>>>> Ph.D Scholar
>>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>>> Ph. No: +91-9897065948
>>>>>>>>
>>>>>>>> ............................................................
>> ...............................................................
>>>>>>>> *A man is the best judge of himself and he has to pay the price for
>>>>>>>> what
>>>>>>>> he
>>>>>>>> does.*
>>>>>>>> ............................................................
>> ...............................................................
>>>>>>>> <students' data after
>>>>>>>> preprocessin.xlsx>_______________________________________________
>>>>>>>> Wekalist mailing list
>>>>>>>> Send posts to: [hidden email]
>>>>>>>> List info and subscription status:
>>>>>>>> https://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: [hidden email]
>>>>>>> List info and subscription status:
>>>>>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>>>>>> List etiquette:
>>>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> Thanks & Regards
>>>>>> Bhupesh Rawat.
>>>>>> Ph.D Scholar
>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>> Ph. No: +91-9897065948
>>>>>>
>>>>>> ............................................................
>> ...............................................................
>>>>>> *A man is the best judge of himself and he has to pay the price for
>> what
>>>>>> he
>>>>>> does.*
>>>>>> ............................................................
>> ...............................................................
>>>>>> _______________________________________________
>>>>>> Wekalist mailing list
>>>>>> Send posts to: [hidden email]
>>>>>> List info and subscription status:
>>>>>> https://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: [hidden email]
>>>>> List info and subscription status:
>>>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>>>> List etiquette:
>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> Thanks & Regards
>>>>> Bhupesh Rawat.
>>>>> Ph.D Scholar
>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>> Ph. No: +91-9897065948
>>>>>
>>>>> ............................................................
>> ...............................................................
>>>>> *A man is the best judge of himself and he has to pay the price for
>> what
>>>>> he
>>>>> does.*
>>>>> ............................................................
>> ...............................................................
>>>>>
>>>>>
>>>>>
>>>>> <knowledge flow
>>>>> interuppted.docx>_______________________________________________
>>>>> Wekalist mailing list
>>>>> Send posts to: [hidden email]
>>>>> List info and subscription status:
>>>>> https://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: [hidden email]
>>>> List info and subscription status:
>>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>>> List etiquette:
>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>
>>>
>>>
>>> --
>>> Thanks & Regards
>>> Bhupesh Rawat.
>>> Ph.D Scholar
>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>> Ph. No: +91-9897065948
>>>
>>> ............................................................
>> ...............................................................
>>> *A man is the best judge of himself and he has to pay the price for what
>> he
>>> does.*
>>> ............................................................
>> ...............................................................
>>> _______________________________________________
>>> Wekalist mailing list
>>> Send posts to: [hidden email]
>>> List info and subscription status: https://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: [hidden email]
>> List info and subscription status: https://list.waikato.ac.nz/
>> mailman/listinfo/wekalist
>> List etiquette: http://www.cs.waikato.ac.nz/~
>> ml/weka/mailinglist_etiquette.html
>>
>
>
>
> --
> Thanks & Regards
> Bhupesh Rawat.
> Ph.D Scholar
> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
> Vidya Vihar,Rai Bareilly road(Lucknow)
> Ph. No: +91-9897065948
>
> ...........................................................................................................................
> *A man is the best judge of himself and he has to pay the price for what he
> does.*
> ...........................................................................................................................
> _______________________________________________
> Wekalist mailing list
> Send posts to: [hidden email]
> List info and subscription status: https://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: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Bhupesh Rawat
This question is related to the implementation of Apriori in MATLAB
which i have been trying to solve for quite some time but with no
positive result. Any help would be highly appreciable. I have attached
a file having two small dataset. the first dataset is running fine
with the Apriori algorithm, however the second dataset almost similar
to the first one except the last row, generates the following error:

??? Attempted to access count.%cell(16); index out of bounds because
numel(count.%cell)=15


% Calculate Patterns Counts

        count{k+1}=zeros(size(C{2}));
        for r=1:numel(C{k+1})
            for i=1:numel(T)
                if IsContainedIn(C{k+1}{r},T{i})
                    count{k+1}(r)=count{k+1}(r)+1;    % line
containing the error
                end
            end
        end



%% Apriori

MST=0.2;   % Minimum Support Threshold

MCT=0.2;    % Minimum Confidence Threshold

[FinalRules, Rules]=Apriori(T,MST,MCT);    % line containing the error

On 3/5/17, Eibe Frank <[hidden email]> wrote:

> You can create a new attribute by combining nominal attributes using the
> CartesianProduct filter.
>
> Regarding the reliability of the rules, take a look at the literature for
> "predictive apriori" on Google Scholar. I don't know if there have been any
> extensive studies.
>
> To get a rough idea of how well PredictiveApriori works for your data,
> regardless of the accuracy of individual rules considered in isolation, you
> could apply it to mine class association rules with the JCBA classifier
> (from the classAssociationRules package) and use cross-validation for
> evaluation, similar to what we did in our paper. Obviously, you will have to
> create an appropriate class attribute for each attribute/attribute
> combination that you are interested in (possibly using CartesianProduct).
>
> Here is an example command-line, running JCBA with PredictiveApriori on the
> vote data (using the default class attribute). I got it to only output the
> top two rules for simplicity:
>
> ===================
>
> java weka.Run .JCBA -A ".PredictiveApriori -N 2" -t ~/datasets/UCI/vote.arff
>
> Options: -A ".PredictiveApriori -N 2"
>
>
> Classification Rules (ordered):
> ==========================
>
> 1. physician-fee-freeze=n 3 0 adoption-of-the-budget-resolution=y 2 1  ==>
> Class=democrat     acc:(0.99),  (219),
> 2. crime=n 13 0 el-salvador-aid=n 4 0 adoption-of-the-budget-resolution=y 2
> 1  ==> Class=democrat     acc:(0.99),  (144),
>
>
> Default Class: Class=republican
>
> Additional Information:
> Number of Classification Associations Rules generated by Rule Miner: 2
> Number of Classification Rules: 2
>
> Mining Time in sec.: 7.867
> Pruning Time in sec. : 0.033
>
>
> Time taken to build model: 7.91 seconds
> Time taken to test model on training data: 0.02 seconds
>
> === Error on training data ===
>
> Correctly Classified Instances         389               89.4253 %
> Incorrectly Classified Instances        46               10.5747 %
> Kappa statistic                          0.7877
> Mean absolute error                      0.1057
> Root mean squared error                  0.3252
> Relative absolute error                 22.2991 %
> Root relative squared error             66.7902 %
> Total Number of Instances              435
>
>
> === Detailed Accuracy By Class ===
>
>                  TP Rate  FP Rate  Precision  Recall   F-Measure  MCC
> ROC Area  PRC Area  Class
>                  0.828    0.000    1.000      0.828    0.906      0.806
> 0.914     0.933     democrat
>                  1.000    0.172    0.785      1.000    0.880      0.806
> 0.914     0.785     republican
> Weighted Avg.    0.894    0.067    0.917      0.894    0.896      0.806
> 0.914     0.876
>
>
> === Confusion Matrix ===
>
>    a   b   <-- classified as
>  221  46 |   a = democrat
>    0 168 |   b = republican
>
>
>
> === Stratified cross-validation ===
>
> Correctly Classified Instances         391               89.8851 %
> Incorrectly Classified Instances        44               10.1149 %
> Kappa statistic                          0.7957
> Mean absolute error                      0.1011
> Root mean squared error                  0.318
> Relative absolute error                 21.3284 %
> Root relative squared error             65.3201 %
> Total Number of Instances              435
>
>
> === Detailed Accuracy By Class ===
>
>                  TP Rate  FP Rate  Precision  Recall   F-Measure  MCC
> ROC Area  PRC Area  Class
>                  0.843    0.012    0.991      0.843    0.911      0.810
> 0.915     0.932     democrat
>                  0.988    0.157    0.798      0.988    0.883      0.810
> 0.915     0.793     republican
> Weighted Avg.    0.899    0.068    0.917      0.899    0.900      0.810
> 0.915     0.878
>
>
> === Confusion Matrix ===
>
>    a   b   <-- classified as
>  225  42 |   a = democrat
>    2 166 |   b = republican
>
> ===================
>
> The observed precision of classifications for class democrat estimated by
> cross-validation (under "Detailed Accuracy By Class") is quite close to the
> accuracy estimates listed for the individual rules in the initial output of
> the class association rules (0.99), so we can be reasonably confident in
> this case that the rules are very accurate.
>
> This process doesn't give you independent accuracy estimates for individual
> rules though. Assuming you have a reasonably large test set, you could code
> up individual rules in PMML and use the PMML classifier in WEKA for each
> rule to evaluate it on the test set
> (http://wiki.pentaho.com/display/DATAMINING/PMML+Support+in+Weka). However,
> you mentioned that you have a small dataset so this is probably not an
> option for you.
>
> Cheers,
> Eibe
>
>> On 5 Mar 2017, at 03:39, Bhupesh Rawat <[hidden email]> wrote:
>>
>> Sir,
>>
>> How to choose combination of attribute as a class attribute with Jrip
>> or PART in weka 3.8.
>>
>> Moreover i tried Predictive apriori on the dataset and as a result i
>> found some rules with their respective accuracy. How reliable are
>> those rules based on this accuracy.
>>
>>
>>
>>
>>
>> On Thu, Mar 2, 2017 at 2:10 AM, Eibe Frank <[hidden email]> wrote:
>>
>>> In WEKA 3.8/3.9, under
>>>
>>>  filters.unsupervised.attribute.CartesianProduct
>>>
>>> Cheers,
>>> Eibe
>>>
>>>> On 1/03/2017, at 6:13 PM, Bhupesh Rawat <[hidden email]> wrote:
>>>>
>>>> Thank you Sir, the problem has been fixed.
>>>>
>>>> Moreover i would also like to use the combination of attributes for
>>>> which  you suggested  the CartesionProduct filter. Where could i find
>>>> this option?
>>>>
>>>> On 2/28/17, Eibe Frank <[hidden email]> wrote:
>>>>> What does the log say (see the “log” tab next to the “status” tab)?
>>>>>
>>>>> Cheers,
>>>>> Eibe
>>>>>
>>>>>> On 27/02/2017, at 11:56 PM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>
>>>>>> Sir,
>>>>>> When I use the KnowledgeFlow GUI the status shown by two of the
>>> components
>>>>>> is interrupted(namely crossvalidationfoldmaker and J48) as shown in
>>>>>> the
>>>>>> attached file. How to fix it?
>>>>>>
>>>>>> On Mon, Feb 27, 2017 at 3:08 AM, Eibe Frank <[hidden email]>
>>> wrote:
>>>>>> In the Explorer, there is no way to turn off evaluation completely.
>>>>>> You
>>>>>> could use the command-line interface or the KnowledgeFlow GUI though.
>>>>>>
>>>>>> Having said this, if you evaluate on the training set, the runtime
>>>>>> overhead is quite small if you apply a rule learner.
>>>>>>
>>>>>> Note also that the Explorer always outputs the classification model
>>>>>> for
>>>>>> the *full* dataset loaded into the Preprocess panel, regardless of
>>> which
>>>>>> evaluation metric you choose, i.e., you will get the rule set for the
>>> full
>>>>>> dataset regardless of the evaluation method you use.
>>>>>>
>>>>>> Cheers,
>>>>>> Eibe
>>>>>>
>>>>>>> On 26/02/2017, at 8:07 PM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>>
>>>>>>> Sir,
>>>>>>>
>>>>>>> How could i perform these two task seperately(applying classification
>>>>>>> rule learner and estimating classification accuracy). The accuracy is
>>>>>>> estimated each time i run the classifier on the dataset.
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> On 2/24/17, Eibe Frank <[hidden email]> wrote:
>>>>>>>> No, not really. However, the dataset is quite small. You could just
>>> run
>>>>>>>> a
>>>>>>>> classification rule learner such as JRip or PART on the data,
>>> treating
>>>>>>>> each
>>>>>>>> of the attributes in turn as the class attribute. Then you can
>>> estimate
>>>>>>>> classification accuracy using cross-validation.
>>>>>>>>
>>>>>>>> You could also create combinations of attributes using the
>>>>>>>> CartesionProduct
>>>>>>>> filter.
>>>>>>>>
>>>>>>>> Cheers,
>>>>>>>> Eibe
>>>>>>>>
>>>>>>>>> On 24/02/2017, at 3:11 AM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>>>>
>>>>>>>>> I have a small dataset which contains student enrolment data in
>>>>>>>>> various courses. If a student has selected a particular course it
>>>>>>>>> is
>>>>>>>>> indicated by ‘Y’ else ‘N’ is used. I have also attached a file for
>>>>>>>>> better understanding of the dataset. I am interested in knowing if
>>> it
>>>>>>>>> is possible to measure the accuracy of the association rules with
>>> this
>>>>>>>>> dataset by the proposed approach in your paper.
>>>>>>>>>
>>>>>>>>> On 2/23/17, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>>>>> Thank you so much for the response!!
>>>>>>>>>> On Feb 23, 2017 8:26 AM, "Eibe Frank" <[hidden email]> wrote:
>>>>>>>>>>
>>>>>>>>>>> You mean beyond confidence, lift, or one the other metrics that
>>> you
>>>>>>>>>>> can
>>>>>>>>>>> get in the output of each rule? This is a tough question. One way
>>>>>>>>>>> may be
>>>>>>>>>>> to
>>>>>>>>>>> use the association rule mining algorithm to build classification
>>>>>>>>>>> rules
>>>>>>>>>>> and
>>>>>>>>>>> then evaluate the accuracy of those classification rules. We had
>>>>>>>>>>> a
>>>>>>>>>>> paper
>>>>>>>>>>> on
>>>>>>>>>>> this quite a while back:
>>>>>>>>>>>
>>>>>>>>>>> Mutter, S., Hall, M., & Frank, E. (2004, December). Using
>>>>>>>>>>> classification
>>>>>>>>>>> to evaluate the output of confidence-based association rule
>>> mining.
>>>>>>>>>>> In
>>>>>>>>>>> Australasian Joint Conference on Artificial Intelligence (pp.
>>>>>>>>>>> 538-549).
>>>>>>>>>>> Springer Berlin Heidelberg.
>>>>>>>>>>>
>>>>>>>>>>> I suppose you could also evaluate the individual association
>>>>>>>>>>> rules
>>>>>>>>>>> on a
>>>>>>>>>>> separate test set, by computing the confidence measure, etc., on
>>> the
>>>>>>>>>>> test
>>>>>>>>>>> set for each rule, but this functionality is not provided by
>>>>>>>>>>> WEKA.
>>>>>>>>>>>
>>>>>>>>>>> Cheers,
>>>>>>>>>>> Eibe
>>>>>>>>>>>
>>>>>>>>>>>> On 23/02/2017, at 12:46 AM, Bhupesh Rawat <[hidden email]>
>>> wrote:
>>>>>>>>>>>>
>>>>>>>>>>>> Dear Sir/Madam
>>>>>>>>>>>>
>>>>>>>>>>>> I have discovered some rules through weka. Could you tell me how
>>> to
>>>>>>>>>>> measure  the accuracy of those rules.
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> On Tue, Feb 14, 2017 at 3:44 AM, Peter Reutemann
>>>>>>>>>>>> <[hidden email]>
>>>>>>>>>>> wrote:
>>>>>>>>>>>>> The size of the final Apriori model as a serialised Java object
>>>>>>>>>>>>> can
>>>>>>>>>>>>> be
>>>>>>>>>>> established saving it to a file and considering the file size.
>>> Note
>>>>>>>>>>> that
>>>>>>>>>>> this is different from the size of the object in memory (see,
>>> e.g.,
>>>>>>>>>>> http://stackoverflow.com/questions/7146559/serialized-
>>>>>>>>>>> object-size-vs-in-memory-object-size-in-java#7146941).
>>>>>>>>>>>>>
>>>>>>>>>>>>> I don’t know of a good way to measure peak memory consumption
>>> of a
>>>>>>>>>>> Java program (after garbage collection). A crude way would be to
>>> run
>>>>>>>>>>> the
>>>>>>>>>>> program from the command-line (to avoid overhead associated with
>>> the
>>>>>>>>>>> GUIs)
>>>>>>>>>>> with different maximum heap sizes, e.g., increasing the heap size
>>>>>>>>>>> until
>>>>>>>>>>> the
>>>>>>>>>>> program runs through. Another option is to look at the heap size
>>> in
>>>>>>>>>>> a
>>>>>>>>>>> profiler (e.g., visualvm), enforcing garbage collection before a
>>>>>>>>>>> readout.
>>>>>>>>>>>>
>>>>>>>>>>>> You can use the sizeofag javaagent for determining the size of a
>>>>>>>>>>>> Java
>>>>>>>>>>> object:
>>>>>>>>>>>> https://github.com/fracpete/sizeofag
>>>>>>>>>>>>
>>>>>>>>>>>> Credits to Maxim Zakharenkov, who wrote the original code.
>>>>>>>>>>>>
>>>>>>>>>>>> Cheers, Peter
>>>>>>>>>>>> --
>>>>>>>>>>>> Peter Reutemann
>>>>>>>>>>>> Dept. of Computer Science
>>>>>>>>>>>> University of Waikato, NZ
>>>>>>>>>>>> +64 (7) 858-5174
>>>>>>>>>>>> http://www.cms.waikato.ac.nz/~fracpete/
>>>>>>>>>>>> http://www.data-mining.co.nz/
>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>> Wekalist mailing list
>>>>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>>>>>>>>> mailman/listinfo/wekalist
>>>>>>>>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>>>>>>>>> ml/weka/mailinglist_etiquette.html
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> --
>>>>>>>>>>>> Thanks & Regards
>>>>>>>>>>>> Bhupesh Rawat.
>>>>>>>>>>>> Ph.D Scholar
>>>>>>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar
>>>>>>>>>>>> University
>>>>>>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>>>>>>> Ph. No: +91-9897065948
>>>>>>>>>>>>
>>>>>>>>>>>> ............................................................
>>>>>>>>>>> ...............................................................
>>>>>>>>>>>> *A man is the best judge of himself and he has to pay the price
>>> for
>>>>>>>>>>>> what
>>>>>>>>>>> he
>>>>>>>>>>>> does.*
>>>>>>>>>>>> ............................................................
>>>>>>>>>>> ...............................................................
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>> Wekalist mailing list
>>>>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>>>>> List info and subscription status: https://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: [hidden email]
>>>>>>>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>>>>>>>>> mailman/listinfo/wekalist
>>>>>>>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>>>>>>>>> ml/weka/mailinglist_etiquette.html
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> --
>>>>>>>>> Thanks & Regards
>>>>>>>>> Bhupesh Rawat.
>>>>>>>>> Ph.D Scholar
>>>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar
>>>>>>>>> University
>>>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>>>> Ph. No: +91-9897065948
>>>>>>>>>
>>>>>>>>> ............................................................
>>> ...............................................................
>>>>>>>>> *A man is the best judge of himself and he has to pay the price for
>>>>>>>>> what
>>>>>>>>> he
>>>>>>>>> does.*
>>>>>>>>> ............................................................
>>> ...............................................................
>>>>>>>>> <students' data after
>>>>>>>>> preprocessin.xlsx>_______________________________________________
>>>>>>>>> Wekalist mailing list
>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>> List info and subscription status:
>>>>>>>>> https://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: [hidden email]
>>>>>>>> List info and subscription status:
>>>>>>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>>>>>>> List etiquette:
>>>>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> --
>>>>>>> Thanks & Regards
>>>>>>> Bhupesh Rawat.
>>>>>>> Ph.D Scholar
>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>> Ph. No: +91-9897065948
>>>>>>>
>>>>>>> ............................................................
>>> ...............................................................
>>>>>>> *A man is the best judge of himself and he has to pay the price for
>>> what
>>>>>>> he
>>>>>>> does.*
>>>>>>> ............................................................
>>> ...............................................................
>>>>>>> _______________________________________________
>>>>>>> Wekalist mailing list
>>>>>>> Send posts to: [hidden email]
>>>>>>> List info and subscription status:
>>>>>>> https://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: [hidden email]
>>>>>> List info and subscription status:
>>>>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>>>>> List etiquette:
>>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> Thanks & Regards
>>>>>> Bhupesh Rawat.
>>>>>> Ph.D Scholar
>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>> Ph. No: +91-9897065948
>>>>>>
>>>>>> ............................................................
>>> ...............................................................
>>>>>> *A man is the best judge of himself and he has to pay the price for
>>> what
>>>>>> he
>>>>>> does.*
>>>>>> ............................................................
>>> ...............................................................
>>>>>>
>>>>>>
>>>>>>
>>>>>> <knowledge flow
>>>>>> interuppted.docx>_______________________________________________
>>>>>> Wekalist mailing list
>>>>>> Send posts to: [hidden email]
>>>>>> List info and subscription status:
>>>>>> https://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: [hidden email]
>>>>> List info and subscription status:
>>>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>>>> List etiquette:
>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>
>>>>
>>>>
>>>> --
>>>> Thanks & Regards
>>>> Bhupesh Rawat.
>>>> Ph.D Scholar
>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>> Ph. No: +91-9897065948
>>>>
>>>> ............................................................
>>> ...............................................................
>>>> *A man is the best judge of himself and he has to pay the price for what
>>> he
>>>> does.*
>>>> ............................................................
>>> ...............................................................
>>>> _______________________________________________
>>>> Wekalist mailing list
>>>> Send posts to: [hidden email]
>>>> List info and subscription status: https://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: [hidden email]
>>> List info and subscription status: https://list.waikato.ac.nz/
>>> mailman/listinfo/wekalist
>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>> ml/weka/mailinglist_etiquette.html
>>>
>>
>>
>>
>> --
>> Thanks & Regards
>> Bhupesh Rawat.
>> Ph.D Scholar
>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>> Vidya Vihar,Rai Bareilly road(Lucknow)
>> Ph. No: +91-9897065948
>>
>> ...........................................................................................................................
>> *A man is the best judge of himself and he has to pay the price for what
>> he
>> does.*
>> ...........................................................................................................................
>> _______________________________________________
>> Wekalist mailing list
>> Send posts to: [hidden email]
>> List info and subscription status:
>> https://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: [hidden email]
> List info and subscription status:
> https://list.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>

--
Thanks & Regards
Bhupesh Rawat.
Ph.D Scholar
Department of Computer Science,Babasaheb Bhimrao Ambedkar University
Vidya Vihar,Rai Bareilly road(Lucknow)
Ph. No: +91-9897065948

...........................................................................................................................
*A man is the best judge of himself and he has to pay the price for what he
does.*
...........................................................................................................................

_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html

dataset_matlab.xls (28K) Download Attachment
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Eibe Frank-2
Administrator
You should post this question in the appropriate help forum for Matlab.

Cheers,
Eibe

> On 6/03/2017, at 7:14 AM, Bhupesh Rawat <[hidden email]> wrote:
>
> This question is related to the implementation of Apriori in MATLAB
> which i have been trying to solve for quite some time but with no
> positive result. Any help would be highly appreciable. I have attached
> a file having two small dataset. the first dataset is running fine
> with the Apriori algorithm, however the second dataset almost similar
> to the first one except the last row, generates the following error:
>
> ??? Attempted to access count.%cell(16); index out of bounds because
> numel(count.%cell)=15
>
>
> % Calculate Patterns Counts
>
>        count{k+1}=zeros(size(C{2}));
>        for r=1:numel(C{k+1})
>            for i=1:numel(T)
>                if IsContainedIn(C{k+1}{r},T{i})
>                    count{k+1}(r)=count{k+1}(r)+1;    % line
> containing the error
>                end
>            end
>        end
>
>
>
> %% Apriori
>
> MST=0.2;   % Minimum Support Threshold
>
> MCT=0.2;    % Minimum Confidence Threshold
>
> [FinalRules, Rules]=Apriori(T,MST,MCT);    % line containing the error
>
> On 3/5/17, Eibe Frank <[hidden email]> wrote:
>> You can create a new attribute by combining nominal attributes using the
>> CartesianProduct filter.
>>
>> Regarding the reliability of the rules, take a look at the literature for
>> "predictive apriori" on Google Scholar. I don't know if there have been any
>> extensive studies.
>>
>> To get a rough idea of how well PredictiveApriori works for your data,
>> regardless of the accuracy of individual rules considered in isolation, you
>> could apply it to mine class association rules with the JCBA classifier
>> (from the classAssociationRules package) and use cross-validation for
>> evaluation, similar to what we did in our paper. Obviously, you will have to
>> create an appropriate class attribute for each attribute/attribute
>> combination that you are interested in (possibly using CartesianProduct).
>>
>> Here is an example command-line, running JCBA with PredictiveApriori on the
>> vote data (using the default class attribute). I got it to only output the
>> top two rules for simplicity:
>>
>> ===================
>>
>> java weka.Run .JCBA -A ".PredictiveApriori -N 2" -t ~/datasets/UCI/vote.arff
>>
>> Options: -A ".PredictiveApriori -N 2"
>>
>>
>> Classification Rules (ordered):
>> ==========================
>>
>> 1. physician-fee-freeze=n 3 0 adoption-of-the-budget-resolution=y 2 1  ==>
>> Class=democrat     acc:(0.99),  (219),
>> 2. crime=n 13 0 el-salvador-aid=n 4 0 adoption-of-the-budget-resolution=y 2
>> 1  ==> Class=democrat     acc:(0.99),  (144),
>>
>>
>> Default Class: Class=republican
>>
>> Additional Information:
>> Number of Classification Associations Rules generated by Rule Miner: 2
>> Number of Classification Rules: 2
>>
>> Mining Time in sec.: 7.867
>> Pruning Time in sec. : 0.033
>>
>>
>> Time taken to build model: 7.91 seconds
>> Time taken to test model on training data: 0.02 seconds
>>
>> === Error on training data ===
>>
>> Correctly Classified Instances         389               89.4253 %
>> Incorrectly Classified Instances        46               10.5747 %
>> Kappa statistic                          0.7877
>> Mean absolute error                      0.1057
>> Root mean squared error                  0.3252
>> Relative absolute error                 22.2991 %
>> Root relative squared error             66.7902 %
>> Total Number of Instances              435
>>
>>
>> === Detailed Accuracy By Class ===
>>
>>                 TP Rate  FP Rate  Precision  Recall   F-Measure  MCC
>> ROC Area  PRC Area  Class
>>                 0.828    0.000    1.000      0.828    0.906      0.806
>> 0.914     0.933     democrat
>>                 1.000    0.172    0.785      1.000    0.880      0.806
>> 0.914     0.785     republican
>> Weighted Avg.    0.894    0.067    0.917      0.894    0.896      0.806
>> 0.914     0.876
>>
>>
>> === Confusion Matrix ===
>>
>>   a   b   <-- classified as
>> 221  46 |   a = democrat
>>   0 168 |   b = republican
>>
>>
>>
>> === Stratified cross-validation ===
>>
>> Correctly Classified Instances         391               89.8851 %
>> Incorrectly Classified Instances        44               10.1149 %
>> Kappa statistic                          0.7957
>> Mean absolute error                      0.1011
>> Root mean squared error                  0.318
>> Relative absolute error                 21.3284 %
>> Root relative squared error             65.3201 %
>> Total Number of Instances              435
>>
>>
>> === Detailed Accuracy By Class ===
>>
>>                 TP Rate  FP Rate  Precision  Recall   F-Measure  MCC
>> ROC Area  PRC Area  Class
>>                 0.843    0.012    0.991      0.843    0.911      0.810
>> 0.915     0.932     democrat
>>                 0.988    0.157    0.798      0.988    0.883      0.810
>> 0.915     0.793     republican
>> Weighted Avg.    0.899    0.068    0.917      0.899    0.900      0.810
>> 0.915     0.878
>>
>>
>> === Confusion Matrix ===
>>
>>   a   b   <-- classified as
>> 225  42 |   a = democrat
>>   2 166 |   b = republican
>>
>> ===================
>>
>> The observed precision of classifications for class democrat estimated by
>> cross-validation (under "Detailed Accuracy By Class") is quite close to the
>> accuracy estimates listed for the individual rules in the initial output of
>> the class association rules (0.99), so we can be reasonably confident in
>> this case that the rules are very accurate.
>>
>> This process doesn't give you independent accuracy estimates for individual
>> rules though. Assuming you have a reasonably large test set, you could code
>> up individual rules in PMML and use the PMML classifier in WEKA for each
>> rule to evaluate it on the test set
>> (http://wiki.pentaho.com/display/DATAMINING/PMML+Support+in+Weka). However,
>> you mentioned that you have a small dataset so this is probably not an
>> option for you.
>>
>> Cheers,
>> Eibe
>>
>>> On 5 Mar 2017, at 03:39, Bhupesh Rawat <[hidden email]> wrote:
>>>
>>> Sir,
>>>
>>> How to choose combination of attribute as a class attribute with Jrip
>>> or PART in weka 3.8.
>>>
>>> Moreover i tried Predictive apriori on the dataset and as a result i
>>> found some rules with their respective accuracy. How reliable are
>>> those rules based on this accuracy.
>>>
>>>
>>>
>>>
>>>
>>> On Thu, Mar 2, 2017 at 2:10 AM, Eibe Frank <[hidden email]> wrote:
>>>
>>>> In WEKA 3.8/3.9, under
>>>>
>>>> filters.unsupervised.attribute.CartesianProduct
>>>>
>>>> Cheers,
>>>> Eibe
>>>>
>>>>> On 1/03/2017, at 6:13 PM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>
>>>>> Thank you Sir, the problem has been fixed.
>>>>>
>>>>> Moreover i would also like to use the combination of attributes for
>>>>> which  you suggested  the CartesionProduct filter. Where could i find
>>>>> this option?
>>>>>
>>>>> On 2/28/17, Eibe Frank <[hidden email]> wrote:
>>>>>> What does the log say (see the “log” tab next to the “status” tab)?
>>>>>>
>>>>>> Cheers,
>>>>>> Eibe
>>>>>>
>>>>>>> On 27/02/2017, at 11:56 PM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>>
>>>>>>> Sir,
>>>>>>> When I use the KnowledgeFlow GUI the status shown by two of the
>>>> components
>>>>>>> is interrupted(namely crossvalidationfoldmaker and J48) as shown in
>>>>>>> the
>>>>>>> attached file. How to fix it?
>>>>>>>
>>>>>>> On Mon, Feb 27, 2017 at 3:08 AM, Eibe Frank <[hidden email]>
>>>> wrote:
>>>>>>> In the Explorer, there is no way to turn off evaluation completely.
>>>>>>> You
>>>>>>> could use the command-line interface or the KnowledgeFlow GUI though.
>>>>>>>
>>>>>>> Having said this, if you evaluate on the training set, the runtime
>>>>>>> overhead is quite small if you apply a rule learner.
>>>>>>>
>>>>>>> Note also that the Explorer always outputs the classification model
>>>>>>> for
>>>>>>> the *full* dataset loaded into the Preprocess panel, regardless of
>>>> which
>>>>>>> evaluation metric you choose, i.e., you will get the rule set for the
>>>> full
>>>>>>> dataset regardless of the evaluation method you use.
>>>>>>>
>>>>>>> Cheers,
>>>>>>> Eibe
>>>>>>>
>>>>>>>> On 26/02/2017, at 8:07 PM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>>>
>>>>>>>> Sir,
>>>>>>>>
>>>>>>>> How could i perform these two task seperately(applying classification
>>>>>>>> rule learner and estimating classification accuracy). The accuracy is
>>>>>>>> estimated each time i run the classifier on the dataset.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> On 2/24/17, Eibe Frank <[hidden email]> wrote:
>>>>>>>>> No, not really. However, the dataset is quite small. You could just
>>>> run
>>>>>>>>> a
>>>>>>>>> classification rule learner such as JRip or PART on the data,
>>>> treating
>>>>>>>>> each
>>>>>>>>> of the attributes in turn as the class attribute. Then you can
>>>> estimate
>>>>>>>>> classification accuracy using cross-validation.
>>>>>>>>>
>>>>>>>>> You could also create combinations of attributes using the
>>>>>>>>> CartesionProduct
>>>>>>>>> filter.
>>>>>>>>>
>>>>>>>>> Cheers,
>>>>>>>>> Eibe
>>>>>>>>>
>>>>>>>>>> On 24/02/2017, at 3:11 AM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>>>>>
>>>>>>>>>> I have a small dataset which contains student enrolment data in
>>>>>>>>>> various courses. If a student has selected a particular course it
>>>>>>>>>> is
>>>>>>>>>> indicated by ‘Y’ else ‘N’ is used. I have also attached a file for
>>>>>>>>>> better understanding of the dataset. I am interested in knowing if
>>>> it
>>>>>>>>>> is possible to measure the accuracy of the association rules with
>>>> this
>>>>>>>>>> dataset by the proposed approach in your paper.
>>>>>>>>>>
>>>>>>>>>> On 2/23/17, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>>>>>> Thank you so much for the response!!
>>>>>>>>>>> On Feb 23, 2017 8:26 AM, "Eibe Frank" <[hidden email]> wrote:
>>>>>>>>>>>
>>>>>>>>>>>> You mean beyond confidence, lift, or one the other metrics that
>>>> you
>>>>>>>>>>>> can
>>>>>>>>>>>> get in the output of each rule? This is a tough question. One way
>>>>>>>>>>>> may be
>>>>>>>>>>>> to
>>>>>>>>>>>> use the association rule mining algorithm to build classification
>>>>>>>>>>>> rules
>>>>>>>>>>>> and
>>>>>>>>>>>> then evaluate the accuracy of those classification rules. We had
>>>>>>>>>>>> a
>>>>>>>>>>>> paper
>>>>>>>>>>>> on
>>>>>>>>>>>> this quite a while back:
>>>>>>>>>>>>
>>>>>>>>>>>> Mutter, S., Hall, M., & Frank, E. (2004, December). Using
>>>>>>>>>>>> classification
>>>>>>>>>>>> to evaluate the output of confidence-based association rule
>>>> mining.
>>>>>>>>>>>> In
>>>>>>>>>>>> Australasian Joint Conference on Artificial Intelligence (pp.
>>>>>>>>>>>> 538-549).
>>>>>>>>>>>> Springer Berlin Heidelberg.
>>>>>>>>>>>>
>>>>>>>>>>>> I suppose you could also evaluate the individual association
>>>>>>>>>>>> rules
>>>>>>>>>>>> on a
>>>>>>>>>>>> separate test set, by computing the confidence measure, etc., on
>>>> the
>>>>>>>>>>>> test
>>>>>>>>>>>> set for each rule, but this functionality is not provided by
>>>>>>>>>>>> WEKA.
>>>>>>>>>>>>
>>>>>>>>>>>> Cheers,
>>>>>>>>>>>> Eibe
>>>>>>>>>>>>
>>>>>>>>>>>>> On 23/02/2017, at 12:46 AM, Bhupesh Rawat <[hidden email]>
>>>> wrote:
>>>>>>>>>>>>>
>>>>>>>>>>>>> Dear Sir/Madam
>>>>>>>>>>>>>
>>>>>>>>>>>>> I have discovered some rules through weka. Could you tell me how
>>>> to
>>>>>>>>>>>> measure  the accuracy of those rules.
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>> On Tue, Feb 14, 2017 at 3:44 AM, Peter Reutemann
>>>>>>>>>>>>> <[hidden email]>
>>>>>>>>>>>> wrote:
>>>>>>>>>>>>>> The size of the final Apriori model as a serialised Java object
>>>>>>>>>>>>>> can
>>>>>>>>>>>>>> be
>>>>>>>>>>>> established saving it to a file and considering the file size.
>>>> Note
>>>>>>>>>>>> that
>>>>>>>>>>>> this is different from the size of the object in memory (see,
>>>> e.g.,
>>>>>>>>>>>> http://stackoverflow.com/questions/7146559/serialized-
>>>>>>>>>>>> object-size-vs-in-memory-object-size-in-java#7146941).
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> I don’t know of a good way to measure peak memory consumption
>>>> of a
>>>>>>>>>>>> Java program (after garbage collection). A crude way would be to
>>>> run
>>>>>>>>>>>> the
>>>>>>>>>>>> program from the command-line (to avoid overhead associated with
>>>> the
>>>>>>>>>>>> GUIs)
>>>>>>>>>>>> with different maximum heap sizes, e.g., increasing the heap size
>>>>>>>>>>>> until
>>>>>>>>>>>> the
>>>>>>>>>>>> program runs through. Another option is to look at the heap size
>>>> in
>>>>>>>>>>>> a
>>>>>>>>>>>> profiler (e.g., visualvm), enforcing garbage collection before a
>>>>>>>>>>>> readout.
>>>>>>>>>>>>>
>>>>>>>>>>>>> You can use the sizeofag javaagent for determining the size of a
>>>>>>>>>>>>> Java
>>>>>>>>>>>> object:
>>>>>>>>>>>>> https://github.com/fracpete/sizeofag
>>>>>>>>>>>>>
>>>>>>>>>>>>> Credits to Maxim Zakharenkov, who wrote the original code.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Cheers, Peter
>>>>>>>>>>>>> --
>>>>>>>>>>>>> Peter Reutemann
>>>>>>>>>>>>> Dept. of Computer Science
>>>>>>>>>>>>> University of Waikato, NZ
>>>>>>>>>>>>> +64 (7) 858-5174
>>>>>>>>>>>>> http://www.cms.waikato.ac.nz/~fracpete/
>>>>>>>>>>>>> http://www.data-mining.co.nz/
>>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>>> Wekalist mailing list
>>>>>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>>>>>>>>>> mailman/listinfo/wekalist
>>>>>>>>>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>>>>>>>>>> ml/weka/mailinglist_etiquette.html
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>> --
>>>>>>>>>>>>> Thanks & Regards
>>>>>>>>>>>>> Bhupesh Rawat.
>>>>>>>>>>>>> Ph.D Scholar
>>>>>>>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar
>>>>>>>>>>>>> University
>>>>>>>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>>>>>>>> Ph. No: +91-9897065948
>>>>>>>>>>>>>
>>>>>>>>>>>>> ............................................................
>>>>>>>>>>>> ...............................................................
>>>>>>>>>>>>> *A man is the best judge of himself and he has to pay the price
>>>> for
>>>>>>>>>>>>> what
>>>>>>>>>>>> he
>>>>>>>>>>>>> does.*
>>>>>>>>>>>>> ............................................................
>>>>>>>>>>>> ...............................................................
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>>> Wekalist mailing list
>>>>>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>>>>>> List info and subscription status: https://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: [hidden email]
>>>>>>>>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>>>>>>>>>> mailman/listinfo/wekalist
>>>>>>>>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>>>>>>>>>> ml/weka/mailinglist_etiquette.html
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> --
>>>>>>>>>> Thanks & Regards
>>>>>>>>>> Bhupesh Rawat.
>>>>>>>>>> Ph.D Scholar
>>>>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar
>>>>>>>>>> University
>>>>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>>>>> Ph. No: +91-9897065948
>>>>>>>>>>
>>>>>>>>>> ............................................................
>>>> ...............................................................
>>>>>>>>>> *A man is the best judge of himself and he has to pay the price for
>>>>>>>>>> what
>>>>>>>>>> he
>>>>>>>>>> does.*
>>>>>>>>>> ............................................................
>>>> ...............................................................
>>>>>>>>>> <students' data after
>>>>>>>>>> preprocessin.xlsx>_______________________________________________
>>>>>>>>>> Wekalist mailing list
>>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>>> List info and subscription status:
>>>>>>>>>> https://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: [hidden email]
>>>>>>>>> List info and subscription status:
>>>>>>>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>>>>>>>> List etiquette:
>>>>>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> --
>>>>>>>> Thanks & Regards
>>>>>>>> Bhupesh Rawat.
>>>>>>>> Ph.D Scholar
>>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>>> Ph. No: +91-9897065948
>>>>>>>>
>>>>>>>> ............................................................
>>>> ...............................................................
>>>>>>>> *A man is the best judge of himself and he has to pay the price for
>>>> what
>>>>>>>> he
>>>>>>>> does.*
>>>>>>>> ............................................................
>>>> ...............................................................
>>>>>>>> _______________________________________________
>>>>>>>> Wekalist mailing list
>>>>>>>> Send posts to: [hidden email]
>>>>>>>> List info and subscription status:
>>>>>>>> https://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: [hidden email]
>>>>>>> List info and subscription status:
>>>>>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>>>>>> List etiquette:
>>>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> --
>>>>>>> Thanks & Regards
>>>>>>> Bhupesh Rawat.
>>>>>>> Ph.D Scholar
>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>> Ph. No: +91-9897065948
>>>>>>>
>>>>>>> ............................................................
>>>> ...............................................................
>>>>>>> *A man is the best judge of himself and he has to pay the price for
>>>> what
>>>>>>> he
>>>>>>> does.*
>>>>>>> ............................................................
>>>> ...............................................................
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> <knowledge flow
>>>>>>> interuppted.docx>_______________________________________________
>>>>>>> Wekalist mailing list
>>>>>>> Send posts to: [hidden email]
>>>>>>> List info and subscription status:
>>>>>>> https://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: [hidden email]
>>>>>> List info and subscription status:
>>>>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>>>>> List etiquette:
>>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> Thanks & Regards
>>>>> Bhupesh Rawat.
>>>>> Ph.D Scholar
>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>> Ph. No: +91-9897065948
>>>>>
>>>>> ............................................................
>>>> ...............................................................
>>>>> *A man is the best judge of himself and he has to pay the price for what
>>>> he
>>>>> does.*
>>>>> ............................................................
>>>> ...............................................................
>>>>> _______________________________________________
>>>>> Wekalist mailing list
>>>>> Send posts to: [hidden email]
>>>>> List info and subscription status: https://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: [hidden email]
>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>> mailman/listinfo/wekalist
>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>> ml/weka/mailinglist_etiquette.html
>>>>
>>>
>>>
>>>
>>> --
>>> Thanks & Regards
>>> Bhupesh Rawat.
>>> Ph.D Scholar
>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>> Ph. No: +91-9897065948
>>>
>>> ...........................................................................................................................
>>> *A man is the best judge of himself and he has to pay the price for what
>>> he
>>> does.*
>>> ...........................................................................................................................
>>> _______________________________________________
>>> Wekalist mailing list
>>> Send posts to: [hidden email]
>>> List info and subscription status:
>>> https://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: [hidden email]
>> List info and subscription status:
>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>> List etiquette:
>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>
>
>
> --
> Thanks & Regards
> Bhupesh Rawat.
> Ph.D Scholar
> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
> Vidya Vihar,Rai Bareilly road(Lucknow)
> Ph. No: +91-9897065948
>
> ...........................................................................................................................
> *A man is the best judge of himself and he has to pay the price for what he
> does.*
> ...........................................................................................................................
> <dataset_matlab.xls>_______________________________________________
> Wekalist mailing list
> Send posts to: [hidden email]
> List info and subscription status: https://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: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Bhupesh Rawat
What are the considerations that one should take into account while
setting the value of support and confidence in Apriori algorithm?



On 3/6/17, Eibe Frank <[hidden email]> wrote:

> You should post this question in the appropriate help forum for Matlab.
>
> Cheers,
> Eibe
>
>> On 6/03/2017, at 7:14 AM, Bhupesh Rawat <[hidden email]> wrote:
>>
>> This question is related to the implementation of Apriori in MATLAB
>> which i have been trying to solve for quite some time but with no
>> positive result. Any help would be highly appreciable. I have attached
>> a file having two small dataset. the first dataset is running fine
>> with the Apriori algorithm, however the second dataset almost similar
>> to the first one except the last row, generates the following error:
>>
>> ??? Attempted to access count.%cell(16); index out of bounds because
>> numel(count.%cell)=15
>>
>>
>> % Calculate Patterns Counts
>>
>>        count{k+1}=zeros(size(C{2}));
>>        for r=1:numel(C{k+1})
>>            for i=1:numel(T)
>>                if IsContainedIn(C{k+1}{r},T{i})
>>                    count{k+1}(r)=count{k+1}(r)+1;    % line
>> containing the error
>>                end
>>            end
>>        end
>>
>>
>>
>> %% Apriori
>>
>> MST=0.2;   % Minimum Support Threshold
>>
>> MCT=0.2;    % Minimum Confidence Threshold
>>
>> [FinalRules, Rules]=Apriori(T,MST,MCT);    % line containing the error
>>
>> On 3/5/17, Eibe Frank <[hidden email]> wrote:
>>> You can create a new attribute by combining nominal attributes using the
>>> CartesianProduct filter.
>>>
>>> Regarding the reliability of the rules, take a look at the literature for
>>> "predictive apriori" on Google Scholar. I don't know if there have been
>>> any
>>> extensive studies.
>>>
>>> To get a rough idea of how well PredictiveApriori works for your data,
>>> regardless of the accuracy of individual rules considered in isolation,
>>> you
>>> could apply it to mine class association rules with the JCBA classifier
>>> (from the classAssociationRules package) and use cross-validation for
>>> evaluation, similar to what we did in our paper. Obviously, you will have
>>> to
>>> create an appropriate class attribute for each attribute/attribute
>>> combination that you are interested in (possibly using CartesianProduct).
>>>
>>> Here is an example command-line, running JCBA with PredictiveApriori on
>>> the
>>> vote data (using the default class attribute). I got it to only output
>>> the
>>> top two rules for simplicity:
>>>
>>> ===================
>>>
>>> java weka.Run .JCBA -A ".PredictiveApriori -N 2" -t
>>> ~/datasets/UCI/vote.arff
>>>
>>> Options: -A ".PredictiveApriori -N 2"
>>>
>>>
>>> Classification Rules (ordered):
>>> ==========================
>>>
>>> 1. physician-fee-freeze=n 3 0 adoption-of-the-budget-resolution=y 2 1
>>> ==>
>>> Class=democrat     acc:(0.99),  (219),
>>> 2. crime=n 13 0 el-salvador-aid=n 4 0 adoption-of-the-budget-resolution=y
>>> 2
>>> 1  ==> Class=democrat     acc:(0.99),  (144),
>>>
>>>
>>> Default Class: Class=republican
>>>
>>> Additional Information:
>>> Number of Classification Associations Rules generated by Rule Miner: 2
>>> Number of Classification Rules: 2
>>>
>>> Mining Time in sec.: 7.867
>>> Pruning Time in sec. : 0.033
>>>
>>>
>>> Time taken to build model: 7.91 seconds
>>> Time taken to test model on training data: 0.02 seconds
>>>
>>> === Error on training data ===
>>>
>>> Correctly Classified Instances         389               89.4253 %
>>> Incorrectly Classified Instances        46               10.5747 %
>>> Kappa statistic                          0.7877
>>> Mean absolute error                      0.1057
>>> Root mean squared error                  0.3252
>>> Relative absolute error                 22.2991 %
>>> Root relative squared error             66.7902 %
>>> Total Number of Instances              435
>>>
>>>
>>> === Detailed Accuracy By Class ===
>>>
>>>                 TP Rate  FP Rate  Precision  Recall   F-Measure  MCC
>>> ROC Area  PRC Area  Class
>>>                 0.828    0.000    1.000      0.828    0.906      0.806
>>> 0.914     0.933     democrat
>>>                 1.000    0.172    0.785      1.000    0.880      0.806
>>> 0.914     0.785     republican
>>> Weighted Avg.    0.894    0.067    0.917      0.894    0.896      0.806
>>> 0.914     0.876
>>>
>>>
>>> === Confusion Matrix ===
>>>
>>>   a   b   <-- classified as
>>> 221  46 |   a = democrat
>>>   0 168 |   b = republican
>>>
>>>
>>>
>>> === Stratified cross-validation ===
>>>
>>> Correctly Classified Instances         391               89.8851 %
>>> Incorrectly Classified Instances        44               10.1149 %
>>> Kappa statistic                          0.7957
>>> Mean absolute error                      0.1011
>>> Root mean squared error                  0.318
>>> Relative absolute error                 21.3284 %
>>> Root relative squared error             65.3201 %
>>> Total Number of Instances              435
>>>
>>>
>>> === Detailed Accuracy By Class ===
>>>
>>>                 TP Rate  FP Rate  Precision  Recall   F-Measure  MCC
>>> ROC Area  PRC Area  Class
>>>                 0.843    0.012    0.991      0.843    0.911      0.810
>>> 0.915     0.932     democrat
>>>                 0.988    0.157    0.798      0.988    0.883      0.810
>>> 0.915     0.793     republican
>>> Weighted Avg.    0.899    0.068    0.917      0.899    0.900      0.810
>>> 0.915     0.878
>>>
>>>
>>> === Confusion Matrix ===
>>>
>>>   a   b   <-- classified as
>>> 225  42 |   a = democrat
>>>   2 166 |   b = republican
>>>
>>> ===================
>>>
>>> The observed precision of classifications for class democrat estimated by
>>> cross-validation (under "Detailed Accuracy By Class") is quite close to
>>> the
>>> accuracy estimates listed for the individual rules in the initial output
>>> of
>>> the class association rules (0.99), so we can be reasonably confident in
>>> this case that the rules are very accurate.
>>>
>>> This process doesn't give you independent accuracy estimates for
>>> individual
>>> rules though. Assuming you have a reasonably large test set, you could
>>> code
>>> up individual rules in PMML and use the PMML classifier in WEKA for each
>>> rule to evaluate it on the test set
>>> (http://wiki.pentaho.com/display/DATAMINING/PMML+Support+in+Weka).
>>> However,
>>> you mentioned that you have a small dataset so this is probably not an
>>> option for you.
>>>
>>> Cheers,
>>> Eibe
>>>
>>>> On 5 Mar 2017, at 03:39, Bhupesh Rawat <[hidden email]> wrote:
>>>>
>>>> Sir,
>>>>
>>>> How to choose combination of attribute as a class attribute with Jrip
>>>> or PART in weka 3.8.
>>>>
>>>> Moreover i tried Predictive apriori on the dataset and as a result i
>>>> found some rules with their respective accuracy. How reliable are
>>>> those rules based on this accuracy.
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> On Thu, Mar 2, 2017 at 2:10 AM, Eibe Frank <[hidden email]> wrote:
>>>>
>>>>> In WEKA 3.8/3.9, under
>>>>>
>>>>> filters.unsupervised.attribute.CartesianProduct
>>>>>
>>>>> Cheers,
>>>>> Eibe
>>>>>
>>>>>> On 1/03/2017, at 6:13 PM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>
>>>>>> Thank you Sir, the problem has been fixed.
>>>>>>
>>>>>> Moreover i would also like to use the combination of attributes for
>>>>>> which  you suggested  the CartesionProduct filter. Where could i find
>>>>>> this option?
>>>>>>
>>>>>> On 2/28/17, Eibe Frank <[hidden email]> wrote:
>>>>>>> What does the log say (see the “log” tab next to the “status” tab)?
>>>>>>>
>>>>>>> Cheers,
>>>>>>> Eibe
>>>>>>>
>>>>>>>> On 27/02/2017, at 11:56 PM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>>>
>>>>>>>> Sir,
>>>>>>>> When I use the KnowledgeFlow GUI the status shown by two of the
>>>>> components
>>>>>>>> is interrupted(namely crossvalidationfoldmaker and J48) as shown in
>>>>>>>> the
>>>>>>>> attached file. How to fix it?
>>>>>>>>
>>>>>>>> On Mon, Feb 27, 2017 at 3:08 AM, Eibe Frank <[hidden email]>
>>>>> wrote:
>>>>>>>> In the Explorer, there is no way to turn off evaluation completely.
>>>>>>>> You
>>>>>>>> could use the command-line interface or the KnowledgeFlow GUI
>>>>>>>> though.
>>>>>>>>
>>>>>>>> Having said this, if you evaluate on the training set, the runtime
>>>>>>>> overhead is quite small if you apply a rule learner.
>>>>>>>>
>>>>>>>> Note also that the Explorer always outputs the classification model
>>>>>>>> for
>>>>>>>> the *full* dataset loaded into the Preprocess panel, regardless of
>>>>> which
>>>>>>>> evaluation metric you choose, i.e., you will get the rule set for
>>>>>>>> the
>>>>> full
>>>>>>>> dataset regardless of the evaluation method you use.
>>>>>>>>
>>>>>>>> Cheers,
>>>>>>>> Eibe
>>>>>>>>
>>>>>>>>> On 26/02/2017, at 8:07 PM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>>>>
>>>>>>>>> Sir,
>>>>>>>>>
>>>>>>>>> How could i perform these two task seperately(applying
>>>>>>>>> classification
>>>>>>>>> rule learner and estimating classification accuracy). The accuracy
>>>>>>>>> is
>>>>>>>>> estimated each time i run the classifier on the dataset.
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> On 2/24/17, Eibe Frank <[hidden email]> wrote:
>>>>>>>>>> No, not really. However, the dataset is quite small. You could
>>>>>>>>>> just
>>>>> run
>>>>>>>>>> a
>>>>>>>>>> classification rule learner such as JRip or PART on the data,
>>>>> treating
>>>>>>>>>> each
>>>>>>>>>> of the attributes in turn as the class attribute. Then you can
>>>>> estimate
>>>>>>>>>> classification accuracy using cross-validation.
>>>>>>>>>>
>>>>>>>>>> You could also create combinations of attributes using the
>>>>>>>>>> CartesionProduct
>>>>>>>>>> filter.
>>>>>>>>>>
>>>>>>>>>> Cheers,
>>>>>>>>>> Eibe
>>>>>>>>>>
>>>>>>>>>>> On 24/02/2017, at 3:11 AM, Bhupesh Rawat <[hidden email]>
>>>>>>>>>>> wrote:
>>>>>>>>>>>
>>>>>>>>>>> I have a small dataset which contains student enrolment data in
>>>>>>>>>>> various courses. If a student has selected a particular course it
>>>>>>>>>>> is
>>>>>>>>>>> indicated by ‘Y’ else ‘N’ is used. I have also attached a file
>>>>>>>>>>> for
>>>>>>>>>>> better understanding of the dataset. I am interested in knowing
>>>>>>>>>>> if
>>>>> it
>>>>>>>>>>> is possible to measure the accuracy of the association rules with
>>>>> this
>>>>>>>>>>> dataset by the proposed approach in your paper.
>>>>>>>>>>>
>>>>>>>>>>> On 2/23/17, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>>>>>>> Thank you so much for the response!!
>>>>>>>>>>>> On Feb 23, 2017 8:26 AM, "Eibe Frank" <[hidden email]>
>>>>>>>>>>>> wrote:
>>>>>>>>>>>>
>>>>>>>>>>>>> You mean beyond confidence, lift, or one the other metrics that
>>>>> you
>>>>>>>>>>>>> can
>>>>>>>>>>>>> get in the output of each rule? This is a tough question. One
>>>>>>>>>>>>> way
>>>>>>>>>>>>> may be
>>>>>>>>>>>>> to
>>>>>>>>>>>>> use the association rule mining algorithm to build
>>>>>>>>>>>>> classification
>>>>>>>>>>>>> rules
>>>>>>>>>>>>> and
>>>>>>>>>>>>> then evaluate the accuracy of those classification rules. We
>>>>>>>>>>>>> had
>>>>>>>>>>>>> a
>>>>>>>>>>>>> paper
>>>>>>>>>>>>> on
>>>>>>>>>>>>> this quite a while back:
>>>>>>>>>>>>>
>>>>>>>>>>>>> Mutter, S., Hall, M., & Frank, E. (2004, December). Using
>>>>>>>>>>>>> classification
>>>>>>>>>>>>> to evaluate the output of confidence-based association rule
>>>>> mining.
>>>>>>>>>>>>> In
>>>>>>>>>>>>> Australasian Joint Conference on Artificial Intelligence (pp.
>>>>>>>>>>>>> 538-549).
>>>>>>>>>>>>> Springer Berlin Heidelberg.
>>>>>>>>>>>>>
>>>>>>>>>>>>> I suppose you could also evaluate the individual association
>>>>>>>>>>>>> rules
>>>>>>>>>>>>> on a
>>>>>>>>>>>>> separate test set, by computing the confidence measure, etc.,
>>>>>>>>>>>>> on
>>>>> the
>>>>>>>>>>>>> test
>>>>>>>>>>>>> set for each rule, but this functionality is not provided by
>>>>>>>>>>>>> WEKA.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Cheers,
>>>>>>>>>>>>> Eibe
>>>>>>>>>>>>>
>>>>>>>>>>>>>> On 23/02/2017, at 12:46 AM, Bhupesh Rawat <[hidden email]>
>>>>> wrote:
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Dear Sir/Madam
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> I have discovered some rules through weka. Could you tell me
>>>>>>>>>>>>>> how
>>>>> to
>>>>>>>>>>>>> measure  the accuracy of those rules.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> On Tue, Feb 14, 2017 at 3:44 AM, Peter Reutemann
>>>>>>>>>>>>>> <[hidden email]>
>>>>>>>>>>>>> wrote:
>>>>>>>>>>>>>>> The size of the final Apriori model as a serialised Java
>>>>>>>>>>>>>>> object
>>>>>>>>>>>>>>> can
>>>>>>>>>>>>>>> be
>>>>>>>>>>>>> established saving it to a file and considering the file size.
>>>>> Note
>>>>>>>>>>>>> that
>>>>>>>>>>>>> this is different from the size of the object in memory (see,
>>>>> e.g.,
>>>>>>>>>>>>> http://stackoverflow.com/questions/7146559/serialized-
>>>>>>>>>>>>> object-size-vs-in-memory-object-size-in-java#7146941).
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> I don’t know of a good way to measure peak memory consumption
>>>>> of a
>>>>>>>>>>>>> Java program (after garbage collection). A crude way would be
>>>>>>>>>>>>> to
>>>>> run
>>>>>>>>>>>>> the
>>>>>>>>>>>>> program from the command-line (to avoid overhead associated
>>>>>>>>>>>>> with
>>>>> the
>>>>>>>>>>>>> GUIs)
>>>>>>>>>>>>> with different maximum heap sizes, e.g., increasing the heap
>>>>>>>>>>>>> size
>>>>>>>>>>>>> until
>>>>>>>>>>>>> the
>>>>>>>>>>>>> program runs through. Another option is to look at the heap
>>>>>>>>>>>>> size
>>>>> in
>>>>>>>>>>>>> a
>>>>>>>>>>>>> profiler (e.g., visualvm), enforcing garbage collection before
>>>>>>>>>>>>> a
>>>>>>>>>>>>> readout.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> You can use the sizeofag javaagent for determining the size of
>>>>>>>>>>>>>> a
>>>>>>>>>>>>>> Java
>>>>>>>>>>>>> object:
>>>>>>>>>>>>>> https://github.com/fracpete/sizeofag
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Credits to Maxim Zakharenkov, who wrote the original code.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Cheers, Peter
>>>>>>>>>>>>>> --
>>>>>>>>>>>>>> Peter Reutemann
>>>>>>>>>>>>>> Dept. of Computer Science
>>>>>>>>>>>>>> University of Waikato, NZ
>>>>>>>>>>>>>> +64 (7) 858-5174
>>>>>>>>>>>>>> http://www.cms.waikato.ac.nz/~fracpete/
>>>>>>>>>>>>>> http://www.data-mining.co.nz/
>>>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>>>> Wekalist mailing list
>>>>>>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>>>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>>>>>>>>>>> mailman/listinfo/wekalist
>>>>>>>>>>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>>>>>>>>>>> ml/weka/mailinglist_etiquette.html
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> --
>>>>>>>>>>>>>> Thanks & Regards
>>>>>>>>>>>>>> Bhupesh Rawat.
>>>>>>>>>>>>>> Ph.D Scholar
>>>>>>>>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar
>>>>>>>>>>>>>> University
>>>>>>>>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>>>>>>>>> Ph. No: +91-9897065948
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> ............................................................
>>>>>>>>>>>>> ...............................................................
>>>>>>>>>>>>>> *A man is the best judge of himself and he has to pay the
>>>>>>>>>>>>>> price
>>>>> for
>>>>>>>>>>>>>> what
>>>>>>>>>>>>> he
>>>>>>>>>>>>>> does.*
>>>>>>>>>>>>>> ............................................................
>>>>>>>>>>>>> ...............................................................
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>>>> Wekalist mailing list
>>>>>>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>>>>>>> List info and subscription status: https://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: [hidden email]
>>>>>>>>>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>>>>>>>>>>> mailman/listinfo/wekalist
>>>>>>>>>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>>>>>>>>>>> ml/weka/mailinglist_etiquette.html
>>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> --
>>>>>>>>>>> Thanks & Regards
>>>>>>>>>>> Bhupesh Rawat.
>>>>>>>>>>> Ph.D Scholar
>>>>>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar
>>>>>>>>>>> University
>>>>>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>>>>>> Ph. No: +91-9897065948
>>>>>>>>>>>
>>>>>>>>>>> ............................................................
>>>>> ...............................................................
>>>>>>>>>>> *A man is the best judge of himself and he has to pay the price
>>>>>>>>>>> for
>>>>>>>>>>> what
>>>>>>>>>>> he
>>>>>>>>>>> does.*
>>>>>>>>>>> ............................................................
>>>>> ...............................................................
>>>>>>>>>>> <students' data after
>>>>>>>>>>> preprocessin.xlsx>_______________________________________________
>>>>>>>>>>> Wekalist mailing list
>>>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>>>> List info and subscription status:
>>>>>>>>>>> https://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: [hidden email]
>>>>>>>>>> List info and subscription status:
>>>>>>>>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>>>>>>>>> List etiquette:
>>>>>>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> --
>>>>>>>>> Thanks & Regards
>>>>>>>>> Bhupesh Rawat.
>>>>>>>>> Ph.D Scholar
>>>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar
>>>>>>>>> University
>>>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>>>> Ph. No: +91-9897065948
>>>>>>>>>
>>>>>>>>> ............................................................
>>>>> ...............................................................
>>>>>>>>> *A man is the best judge of himself and he has to pay the price for
>>>>> what
>>>>>>>>> he
>>>>>>>>> does.*
>>>>>>>>> ............................................................
>>>>> ...............................................................
>>>>>>>>> _______________________________________________
>>>>>>>>> Wekalist mailing list
>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>> List info and subscription status:
>>>>>>>>> https://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: [hidden email]
>>>>>>>> List info and subscription status:
>>>>>>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>>>>>>> List etiquette:
>>>>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> --
>>>>>>>> Thanks & Regards
>>>>>>>> Bhupesh Rawat.
>>>>>>>> Ph.D Scholar
>>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>>> Ph. No: +91-9897065948
>>>>>>>>
>>>>>>>> ............................................................
>>>>> ...............................................................
>>>>>>>> *A man is the best judge of himself and he has to pay the price for
>>>>> what
>>>>>>>> he
>>>>>>>> does.*
>>>>>>>> ............................................................
>>>>> ...............................................................
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> <knowledge flow
>>>>>>>> interuppted.docx>_______________________________________________
>>>>>>>> Wekalist mailing list
>>>>>>>> Send posts to: [hidden email]
>>>>>>>> List info and subscription status:
>>>>>>>> https://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: [hidden email]
>>>>>>> List info and subscription status:
>>>>>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>>>>>> List etiquette:
>>>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> Thanks & Regards
>>>>>> Bhupesh Rawat.
>>>>>> Ph.D Scholar
>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>> Ph. No: +91-9897065948
>>>>>>
>>>>>> ............................................................
>>>>> ...............................................................
>>>>>> *A man is the best judge of himself and he has to pay the price for
>>>>>> what
>>>>> he
>>>>>> does.*
>>>>>> ............................................................
>>>>> ...............................................................
>>>>>> _______________________________________________
>>>>>> Wekalist mailing list
>>>>>> Send posts to: [hidden email]
>>>>>> List info and subscription status: https://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: [hidden email]
>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>>> mailman/listinfo/wekalist
>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>>> ml/weka/mailinglist_etiquette.html
>>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> Thanks & Regards
>>>> Bhupesh Rawat.
>>>> Ph.D Scholar
>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>> Ph. No: +91-9897065948
>>>>
>>>> ...........................................................................................................................
>>>> *A man is the best judge of himself and he has to pay the price for what
>>>> he
>>>> does.*
>>>> ...........................................................................................................................
>>>> _______________________________________________
>>>> Wekalist mailing list
>>>> Send posts to: [hidden email]
>>>> List info and subscription status:
>>>> https://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: [hidden email]
>>> List info and subscription status:
>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>> List etiquette:
>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>
>>
>>
>> --
>> Thanks & Regards
>> Bhupesh Rawat.
>> Ph.D Scholar
>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>> Vidya Vihar,Rai Bareilly road(Lucknow)
>> Ph. No: +91-9897065948
>>
>> ...........................................................................................................................
>> *A man is the best judge of himself and he has to pay the price for what
>> he
>> does.*
>> ...........................................................................................................................
>> <dataset_matlab.xls>_______________________________________________
>> Wekalist mailing list
>> Send posts to: [hidden email]
>> List info and subscription status:
>> https://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: [hidden email]
> List info and subscription status:
> https://list.waikato.ac.nz/mailman/listinfo/wekalist
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>


--
Thanks & Regards
Bhupesh Rawat.
Ph.D Scholar
Department of Computer Science,Babasaheb Bhimrao Ambedkar University
Vidya Vihar,Rai Bareilly road(Lucknow)
Ph. No: +91-9897065948

...........................................................................................................................
*A man is the best judge of himself and he has to pay the price for what he
does.*
...........................................................................................................................
_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Measuring accuracy and efficiency of association rules!

Eibe Frank-2
Administrator
That will depend on your application, more specifically, the minimum accuracy that you want the rules to achieve and the minimum amount of data that should support each rule.

Note that Apriori is primarily a tool for exploratory analysis and the main goal is normally to identify "interesting" rules. Interestingness may not be directly related to accuracy. Also, often, interesting rules have quite limited support. Unfortunately, lowering the support to a small value so that these rules can be captured can hugely increase the total number of rules found and the runtime of the algorithm.

Cheers,
Eibe

> On 7 Mar 2017, at 06:29, Bhupesh Rawat <[hidden email]> wrote:
>
> What are the considerations that one should take into account while
> setting the value of support and confidence in Apriori algorithm?
>
>
>
> On 3/6/17, Eibe Frank <[hidden email]> wrote:
>> You should post this question in the appropriate help forum for Matlab.
>>
>> Cheers,
>> Eibe
>>
>>> On 6/03/2017, at 7:14 AM, Bhupesh Rawat <[hidden email]> wrote:
>>>
>>> This question is related to the implementation of Apriori in MATLAB
>>> which i have been trying to solve for quite some time but with no
>>> positive result. Any help would be highly appreciable. I have attached
>>> a file having two small dataset. the first dataset is running fine
>>> with the Apriori algorithm, however the second dataset almost similar
>>> to the first one except the last row, generates the following error:
>>>
>>> ??? Attempted to access count.%cell(16); index out of bounds because
>>> numel(count.%cell)=15
>>>
>>>
>>> % Calculate Patterns Counts
>>>
>>>       count{k+1}=zeros(size(C{2}));
>>>       for r=1:numel(C{k+1})
>>>           for i=1:numel(T)
>>>               if IsContainedIn(C{k+1}{r},T{i})
>>>                   count{k+1}(r)=count{k+1}(r)+1;    % line
>>> containing the error
>>>               end
>>>           end
>>>       end
>>>
>>>
>>>
>>> %% Apriori
>>>
>>> MST=0.2;   % Minimum Support Threshold
>>>
>>> MCT=0.2;    % Minimum Confidence Threshold
>>>
>>> [FinalRules, Rules]=Apriori(T,MST,MCT);    % line containing the error
>>>
>>> On 3/5/17, Eibe Frank <[hidden email]> wrote:
>>>> You can create a new attribute by combining nominal attributes using the
>>>> CartesianProduct filter.
>>>>
>>>> Regarding the reliability of the rules, take a look at the literature for
>>>> "predictive apriori" on Google Scholar. I don't know if there have been
>>>> any
>>>> extensive studies.
>>>>
>>>> To get a rough idea of how well PredictiveApriori works for your data,
>>>> regardless of the accuracy of individual rules considered in isolation,
>>>> you
>>>> could apply it to mine class association rules with the JCBA classifier
>>>> (from the classAssociationRules package) and use cross-validation for
>>>> evaluation, similar to what we did in our paper. Obviously, you will have
>>>> to
>>>> create an appropriate class attribute for each attribute/attribute
>>>> combination that you are interested in (possibly using CartesianProduct).
>>>>
>>>> Here is an example command-line, running JCBA with PredictiveApriori on
>>>> the
>>>> vote data (using the default class attribute). I got it to only output
>>>> the
>>>> top two rules for simplicity:
>>>>
>>>> ===================
>>>>
>>>> java weka.Run .JCBA -A ".PredictiveApriori -N 2" -t
>>>> ~/datasets/UCI/vote.arff
>>>>
>>>> Options: -A ".PredictiveApriori -N 2"
>>>>
>>>>
>>>> Classification Rules (ordered):
>>>> ==========================
>>>>
>>>> 1. physician-fee-freeze=n 3 0 adoption-of-the-budget-resolution=y 2 1
>>>> ==>
>>>> Class=democrat     acc:(0.99),  (219),
>>>> 2. crime=n 13 0 el-salvador-aid=n 4 0 adoption-of-the-budget-resolution=y
>>>> 2
>>>> 1  ==> Class=democrat     acc:(0.99),  (144),
>>>>
>>>>
>>>> Default Class: Class=republican
>>>>
>>>> Additional Information:
>>>> Number of Classification Associations Rules generated by Rule Miner: 2
>>>> Number of Classification Rules: 2
>>>>
>>>> Mining Time in sec.: 7.867
>>>> Pruning Time in sec. : 0.033
>>>>
>>>>
>>>> Time taken to build model: 7.91 seconds
>>>> Time taken to test model on training data: 0.02 seconds
>>>>
>>>> === Error on training data ===
>>>>
>>>> Correctly Classified Instances         389               89.4253 %
>>>> Incorrectly Classified Instances        46               10.5747 %
>>>> Kappa statistic                          0.7877
>>>> Mean absolute error                      0.1057
>>>> Root mean squared error                  0.3252
>>>> Relative absolute error                 22.2991 %
>>>> Root relative squared error             66.7902 %
>>>> Total Number of Instances              435
>>>>
>>>>
>>>> === Detailed Accuracy By Class ===
>>>>
>>>>                TP Rate  FP Rate  Precision  Recall   F-Measure  MCC
>>>> ROC Area  PRC Area  Class
>>>>                0.828    0.000    1.000      0.828    0.906      0.806
>>>> 0.914     0.933     democrat
>>>>                1.000    0.172    0.785      1.000    0.880      0.806
>>>> 0.914     0.785     republican
>>>> Weighted Avg.    0.894    0.067    0.917      0.894    0.896      0.806
>>>> 0.914     0.876
>>>>
>>>>
>>>> === Confusion Matrix ===
>>>>
>>>>  a   b   <-- classified as
>>>> 221  46 |   a = democrat
>>>>  0 168 |   b = republican
>>>>
>>>>
>>>>
>>>> === Stratified cross-validation ===
>>>>
>>>> Correctly Classified Instances         391               89.8851 %
>>>> Incorrectly Classified Instances        44               10.1149 %
>>>> Kappa statistic                          0.7957
>>>> Mean absolute error                      0.1011
>>>> Root mean squared error                  0.318
>>>> Relative absolute error                 21.3284 %
>>>> Root relative squared error             65.3201 %
>>>> Total Number of Instances              435
>>>>
>>>>
>>>> === Detailed Accuracy By Class ===
>>>>
>>>>                TP Rate  FP Rate  Precision  Recall   F-Measure  MCC
>>>> ROC Area  PRC Area  Class
>>>>                0.843    0.012    0.991      0.843    0.911      0.810
>>>> 0.915     0.932     democrat
>>>>                0.988    0.157    0.798      0.988    0.883      0.810
>>>> 0.915     0.793     republican
>>>> Weighted Avg.    0.899    0.068    0.917      0.899    0.900      0.810
>>>> 0.915     0.878
>>>>
>>>>
>>>> === Confusion Matrix ===
>>>>
>>>>  a   b   <-- classified as
>>>> 225  42 |   a = democrat
>>>>  2 166 |   b = republican
>>>>
>>>> ===================
>>>>
>>>> The observed precision of classifications for class democrat estimated by
>>>> cross-validation (under "Detailed Accuracy By Class") is quite close to
>>>> the
>>>> accuracy estimates listed for the individual rules in the initial output
>>>> of
>>>> the class association rules (0.99), so we can be reasonably confident in
>>>> this case that the rules are very accurate.
>>>>
>>>> This process doesn't give you independent accuracy estimates for
>>>> individual
>>>> rules though. Assuming you have a reasonably large test set, you could
>>>> code
>>>> up individual rules in PMML and use the PMML classifier in WEKA for each
>>>> rule to evaluate it on the test set
>>>> (http://wiki.pentaho.com/display/DATAMINING/PMML+Support+in+Weka).
>>>> However,
>>>> you mentioned that you have a small dataset so this is probably not an
>>>> option for you.
>>>>
>>>> Cheers,
>>>> Eibe
>>>>
>>>>> On 5 Mar 2017, at 03:39, Bhupesh Rawat <[hidden email]> wrote:
>>>>>
>>>>> Sir,
>>>>>
>>>>> How to choose combination of attribute as a class attribute with Jrip
>>>>> or PART in weka 3.8.
>>>>>
>>>>> Moreover i tried Predictive apriori on the dataset and as a result i
>>>>> found some rules with their respective accuracy. How reliable are
>>>>> those rules based on this accuracy.
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> On Thu, Mar 2, 2017 at 2:10 AM, Eibe Frank <[hidden email]> wrote:
>>>>>
>>>>>> In WEKA 3.8/3.9, under
>>>>>>
>>>>>> filters.unsupervised.attribute.CartesianProduct
>>>>>>
>>>>>> Cheers,
>>>>>> Eibe
>>>>>>
>>>>>>> On 1/03/2017, at 6:13 PM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>>
>>>>>>> Thank you Sir, the problem has been fixed.
>>>>>>>
>>>>>>> Moreover i would also like to use the combination of attributes for
>>>>>>> which  you suggested  the CartesionProduct filter. Where could i find
>>>>>>> this option?
>>>>>>>
>>>>>>> On 2/28/17, Eibe Frank <[hidden email]> wrote:
>>>>>>>> What does the log say (see the “log” tab next to the “status” tab)?
>>>>>>>>
>>>>>>>> Cheers,
>>>>>>>> Eibe
>>>>>>>>
>>>>>>>>> On 27/02/2017, at 11:56 PM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>>>>
>>>>>>>>> Sir,
>>>>>>>>> When I use the KnowledgeFlow GUI the status shown by two of the
>>>>>> components
>>>>>>>>> is interrupted(namely crossvalidationfoldmaker and J48) as shown in
>>>>>>>>> the
>>>>>>>>> attached file. How to fix it?
>>>>>>>>>
>>>>>>>>> On Mon, Feb 27, 2017 at 3:08 AM, Eibe Frank <[hidden email]>
>>>>>> wrote:
>>>>>>>>> In the Explorer, there is no way to turn off evaluation completely.
>>>>>>>>> You
>>>>>>>>> could use the command-line interface or the KnowledgeFlow GUI
>>>>>>>>> though.
>>>>>>>>>
>>>>>>>>> Having said this, if you evaluate on the training set, the runtime
>>>>>>>>> overhead is quite small if you apply a rule learner.
>>>>>>>>>
>>>>>>>>> Note also that the Explorer always outputs the classification model
>>>>>>>>> for
>>>>>>>>> the *full* dataset loaded into the Preprocess panel, regardless of
>>>>>> which
>>>>>>>>> evaluation metric you choose, i.e., you will get the rule set for
>>>>>>>>> the
>>>>>> full
>>>>>>>>> dataset regardless of the evaluation method you use.
>>>>>>>>>
>>>>>>>>> Cheers,
>>>>>>>>> Eibe
>>>>>>>>>
>>>>>>>>>> On 26/02/2017, at 8:07 PM, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>>>>>
>>>>>>>>>> Sir,
>>>>>>>>>>
>>>>>>>>>> How could i perform these two task seperately(applying
>>>>>>>>>> classification
>>>>>>>>>> rule learner and estimating classification accuracy). The accuracy
>>>>>>>>>> is
>>>>>>>>>> estimated each time i run the classifier on the dataset.
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> On 2/24/17, Eibe Frank <[hidden email]> wrote:
>>>>>>>>>>> No, not really. However, the dataset is quite small. You could
>>>>>>>>>>> just
>>>>>> run
>>>>>>>>>>> a
>>>>>>>>>>> classification rule learner such as JRip or PART on the data,
>>>>>> treating
>>>>>>>>>>> each
>>>>>>>>>>> of the attributes in turn as the class attribute. Then you can
>>>>>> estimate
>>>>>>>>>>> classification accuracy using cross-validation.
>>>>>>>>>>>
>>>>>>>>>>> You could also create combinations of attributes using the
>>>>>>>>>>> CartesionProduct
>>>>>>>>>>> filter.
>>>>>>>>>>>
>>>>>>>>>>> Cheers,
>>>>>>>>>>> Eibe
>>>>>>>>>>>
>>>>>>>>>>>> On 24/02/2017, at 3:11 AM, Bhupesh Rawat <[hidden email]>
>>>>>>>>>>>> wrote:
>>>>>>>>>>>>
>>>>>>>>>>>> I have a small dataset which contains student enrolment data in
>>>>>>>>>>>> various courses. If a student has selected a particular course it
>>>>>>>>>>>> is
>>>>>>>>>>>> indicated by ‘Y’ else ‘N’ is used. I have also attached a file
>>>>>>>>>>>> for
>>>>>>>>>>>> better understanding of the dataset. I am interested in knowing
>>>>>>>>>>>> if
>>>>>> it
>>>>>>>>>>>> is possible to measure the accuracy of the association rules with
>>>>>> this
>>>>>>>>>>>> dataset by the proposed approach in your paper.
>>>>>>>>>>>>
>>>>>>>>>>>> On 2/23/17, Bhupesh Rawat <[hidden email]> wrote:
>>>>>>>>>>>>> Thank you so much for the response!!
>>>>>>>>>>>>> On Feb 23, 2017 8:26 AM, "Eibe Frank" <[hidden email]>
>>>>>>>>>>>>> wrote:
>>>>>>>>>>>>>
>>>>>>>>>>>>>> You mean beyond confidence, lift, or one the other metrics that
>>>>>> you
>>>>>>>>>>>>>> can
>>>>>>>>>>>>>> get in the output of each rule? This is a tough question. One
>>>>>>>>>>>>>> way
>>>>>>>>>>>>>> may be
>>>>>>>>>>>>>> to
>>>>>>>>>>>>>> use the association rule mining algorithm to build
>>>>>>>>>>>>>> classification
>>>>>>>>>>>>>> rules
>>>>>>>>>>>>>> and
>>>>>>>>>>>>>> then evaluate the accuracy of those classification rules. We
>>>>>>>>>>>>>> had
>>>>>>>>>>>>>> a
>>>>>>>>>>>>>> paper
>>>>>>>>>>>>>> on
>>>>>>>>>>>>>> this quite a while back:
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Mutter, S., Hall, M., & Frank, E. (2004, December). Using
>>>>>>>>>>>>>> classification
>>>>>>>>>>>>>> to evaluate the output of confidence-based association rule
>>>>>> mining.
>>>>>>>>>>>>>> In
>>>>>>>>>>>>>> Australasian Joint Conference on Artificial Intelligence (pp.
>>>>>>>>>>>>>> 538-549).
>>>>>>>>>>>>>> Springer Berlin Heidelberg.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> I suppose you could also evaluate the individual association
>>>>>>>>>>>>>> rules
>>>>>>>>>>>>>> on a
>>>>>>>>>>>>>> separate test set, by computing the confidence measure, etc.,
>>>>>>>>>>>>>> on
>>>>>> the
>>>>>>>>>>>>>> test
>>>>>>>>>>>>>> set for each rule, but this functionality is not provided by
>>>>>>>>>>>>>> WEKA.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Cheers,
>>>>>>>>>>>>>> Eibe
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> On 23/02/2017, at 12:46 AM, Bhupesh Rawat <[hidden email]>
>>>>>> wrote:
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Dear Sir/Madam
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> I have discovered some rules through weka. Could you tell me
>>>>>>>>>>>>>>> how
>>>>>> to
>>>>>>>>>>>>>> measure  the accuracy of those rules.
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> On Tue, Feb 14, 2017 at 3:44 AM, Peter Reutemann
>>>>>>>>>>>>>>> <[hidden email]>
>>>>>>>>>>>>>> wrote:
>>>>>>>>>>>>>>>> The size of the final Apriori model as a serialised Java
>>>>>>>>>>>>>>>> object
>>>>>>>>>>>>>>>> can
>>>>>>>>>>>>>>>> be
>>>>>>>>>>>>>> established saving it to a file and considering the file size.
>>>>>> Note
>>>>>>>>>>>>>> that
>>>>>>>>>>>>>> this is different from the size of the object in memory (see,
>>>>>> e.g.,
>>>>>>>>>>>>>> http://stackoverflow.com/questions/7146559/serialized-
>>>>>>>>>>>>>> object-size-vs-in-memory-object-size-in-java#7146941).
>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>> I don’t know of a good way to measure peak memory consumption
>>>>>> of a
>>>>>>>>>>>>>> Java program (after garbage collection). A crude way would be
>>>>>>>>>>>>>> to
>>>>>> run
>>>>>>>>>>>>>> the
>>>>>>>>>>>>>> program from the command-line (to avoid overhead associated
>>>>>>>>>>>>>> with
>>>>>> the
>>>>>>>>>>>>>> GUIs)
>>>>>>>>>>>>>> with different maximum heap sizes, e.g., increasing the heap
>>>>>>>>>>>>>> size
>>>>>>>>>>>>>> until
>>>>>>>>>>>>>> the
>>>>>>>>>>>>>> program runs through. Another option is to look at the heap
>>>>>>>>>>>>>> size
>>>>>> in
>>>>>>>>>>>>>> a
>>>>>>>>>>>>>> profiler (e.g., visualvm), enforcing garbage collection before
>>>>>>>>>>>>>> a
>>>>>>>>>>>>>> readout.
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> You can use the sizeofag javaagent for determining the size of
>>>>>>>>>>>>>>> a
>>>>>>>>>>>>>>> Java
>>>>>>>>>>>>>> object:
>>>>>>>>>>>>>>> https://github.com/fracpete/sizeofag
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Credits to Maxim Zakharenkov, who wrote the original code.
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Cheers, Peter
>>>>>>>>>>>>>>> --
>>>>>>>>>>>>>>> Peter Reutemann
>>>>>>>>>>>>>>> Dept. of Computer Science
>>>>>>>>>>>>>>> University of Waikato, NZ
>>>>>>>>>>>>>>> +64 (7) 858-5174
>>>>>>>>>>>>>>> http://www.cms.waikato.ac.nz/~fracpete/
>>>>>>>>>>>>>>> http://www.data-mining.co.nz/
>>>>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>>>>> Wekalist mailing list
>>>>>>>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>>>>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>>>>>>>>>>>> mailman/listinfo/wekalist
>>>>>>>>>>>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>>>>>>>>>>>> ml/weka/mailinglist_etiquette.html
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> --
>>>>>>>>>>>>>>> Thanks & Regards
>>>>>>>>>>>>>>> Bhupesh Rawat.
>>>>>>>>>>>>>>> Ph.D Scholar
>>>>>>>>>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar
>>>>>>>>>>>>>>> University
>>>>>>>>>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>>>>>>>>>> Ph. No: +91-9897065948
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> ............................................................
>>>>>>>>>>>>>> ...............................................................
>>>>>>>>>>>>>>> *A man is the best judge of himself and he has to pay the
>>>>>>>>>>>>>>> price
>>>>>> for
>>>>>>>>>>>>>>> what
>>>>>>>>>>>>>> he
>>>>>>>>>>>>>>> does.*
>>>>>>>>>>>>>>> ............................................................
>>>>>>>>>>>>>> ...............................................................
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>>>>> Wekalist mailing list
>>>>>>>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>>>>>>>> List info and subscription status: https://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: [hidden email]
>>>>>>>>>>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>>>>>>>>>>>> mailman/listinfo/wekalist
>>>>>>>>>>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>>>>>>>>>>>> ml/weka/mailinglist_etiquette.html
>>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> --
>>>>>>>>>>>> Thanks & Regards
>>>>>>>>>>>> Bhupesh Rawat.
>>>>>>>>>>>> Ph.D Scholar
>>>>>>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar
>>>>>>>>>>>> University
>>>>>>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>>>>>>> Ph. No: +91-9897065948
>>>>>>>>>>>>
>>>>>>>>>>>> ............................................................
>>>>>> ...............................................................
>>>>>>>>>>>> *A man is the best judge of himself and he has to pay the price
>>>>>>>>>>>> for
>>>>>>>>>>>> what
>>>>>>>>>>>> he
>>>>>>>>>>>> does.*
>>>>>>>>>>>> ............................................................
>>>>>> ...............................................................
>>>>>>>>>>>> <students' data after
>>>>>>>>>>>> preprocessin.xlsx>_______________________________________________
>>>>>>>>>>>> Wekalist mailing list
>>>>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>>>>> List info and subscription status:
>>>>>>>>>>>> https://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: [hidden email]
>>>>>>>>>>> List info and subscription status:
>>>>>>>>>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>>>>>>>>>> List etiquette:
>>>>>>>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> --
>>>>>>>>>> Thanks & Regards
>>>>>>>>>> Bhupesh Rawat.
>>>>>>>>>> Ph.D Scholar
>>>>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar
>>>>>>>>>> University
>>>>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>>>>> Ph. No: +91-9897065948
>>>>>>>>>>
>>>>>>>>>> ............................................................
>>>>>> ...............................................................
>>>>>>>>>> *A man is the best judge of himself and he has to pay the price for
>>>>>> what
>>>>>>>>>> he
>>>>>>>>>> does.*
>>>>>>>>>> ............................................................
>>>>>> ...............................................................
>>>>>>>>>> _______________________________________________
>>>>>>>>>> Wekalist mailing list
>>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>>> List info and subscription status:
>>>>>>>>>> https://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: [hidden email]
>>>>>>>>> List info and subscription status:
>>>>>>>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>>>>>>>> List etiquette:
>>>>>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> --
>>>>>>>>> Thanks & Regards
>>>>>>>>> Bhupesh Rawat.
>>>>>>>>> Ph.D Scholar
>>>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>>>> Ph. No: +91-9897065948
>>>>>>>>>
>>>>>>>>> ............................................................
>>>>>> ...............................................................
>>>>>>>>> *A man is the best judge of himself and he has to pay the price for
>>>>>> what
>>>>>>>>> he
>>>>>>>>> does.*
>>>>>>>>> ............................................................
>>>>>> ...............................................................
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> <knowledge flow
>>>>>>>>> interuppted.docx>_______________________________________________
>>>>>>>>> Wekalist mailing list
>>>>>>>>> Send posts to: [hidden email]
>>>>>>>>> List info and subscription status:
>>>>>>>>> https://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: [hidden email]
>>>>>>>> List info and subscription status:
>>>>>>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>>>>>>> List etiquette:
>>>>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> --
>>>>>>> Thanks & Regards
>>>>>>> Bhupesh Rawat.
>>>>>>> Ph.D Scholar
>>>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>>>> Ph. No: +91-9897065948
>>>>>>>
>>>>>>> ............................................................
>>>>>> ...............................................................
>>>>>>> *A man is the best judge of himself and he has to pay the price for
>>>>>>> what
>>>>>> he
>>>>>>> does.*
>>>>>>> ............................................................
>>>>>> ...............................................................
>>>>>>> _______________________________________________
>>>>>>> Wekalist mailing list
>>>>>>> Send posts to: [hidden email]
>>>>>>> List info and subscription status: https://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: [hidden email]
>>>>>> List info and subscription status: https://list.waikato.ac.nz/
>>>>>> mailman/listinfo/wekalist
>>>>>> List etiquette: http://www.cs.waikato.ac.nz/~
>>>>>> ml/weka/mailinglist_etiquette.html
>>>>>>
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> Thanks & Regards
>>>>> Bhupesh Rawat.
>>>>> Ph.D Scholar
>>>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>>>> Ph. No: +91-9897065948
>>>>>
>>>>> ...........................................................................................................................
>>>>> *A man is the best judge of himself and he has to pay the price for what
>>>>> he
>>>>> does.*
>>>>> ...........................................................................................................................
>>>>> _______________________________________________
>>>>> Wekalist mailing list
>>>>> Send posts to: [hidden email]
>>>>> List info and subscription status:
>>>>> https://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: [hidden email]
>>>> List info and subscription status:
>>>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>>>> List etiquette:
>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>
>>>
>>>
>>> --
>>> Thanks & Regards
>>> Bhupesh Rawat.
>>> Ph.D Scholar
>>> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
>>> Vidya Vihar,Rai Bareilly road(Lucknow)
>>> Ph. No: +91-9897065948
>>>
>>> ...........................................................................................................................
>>> *A man is the best judge of himself and he has to pay the price for what
>>> he
>>> does.*
>>> ...........................................................................................................................
>>> <dataset_matlab.xls>_______________________________________________
>>> Wekalist mailing list
>>> Send posts to: [hidden email]
>>> List info and subscription status:
>>> https://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: [hidden email]
>> List info and subscription status:
>> https://list.waikato.ac.nz/mailman/listinfo/wekalist
>> List etiquette:
>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>
>
>
> --
> Thanks & Regards
> Bhupesh Rawat.
> Ph.D Scholar
> Department of Computer Science,Babasaheb Bhimrao Ambedkar University
> Vidya Vihar,Rai Bareilly road(Lucknow)
> Ph. No: +91-9897065948
>
> ...........................................................................................................................
> *A man is the best judge of himself and he has to pay the price for what he
> does.*
> ...........................................................................................................................
> _______________________________________________
> Wekalist mailing list
> Send posts to: [hidden email]
> List info and subscription status: https://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: [hidden email]
List info and subscription status: https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
123
Loading...