CorrelationAttributeEval

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

CorrelationAttributeEval

hudabenhar

Hello everyone,
I'm using the CorrelationAttributeEval ranking technique in a classification problem. The dataset I'm working on has numerical and nominal attributes in addition to the class variable(nominal).
To my knowledge CorrelationAttributeEval is based on Pearson's correlation coefficient which only works with continuous variables.
I want to understand how does CorrelationAttributeEval work in this case. Are the numeric features discretized and the symmetrical uncertainty measure used to calculate the correlation between the attributes and the class variable as it is the case for CFS?


_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
To subscribe, unsubscribe, etc., visit 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
|

Re: CorrelationAttributeEval

Eibe Frank-2
Administrator

There is some relevant information to be found under “More”:

 

“Nominal attributes are considered on a value by value basis by treating each value as an indicator. An overall correlation for a nominal attribute is arrived at via a weighted average.”

 

Looking at the code (https://svn.cms.waikato.ac.nz/svn/weka/trunk/weka/src/main/java/weka/attributeSelection/CorrelationAttributeEval.java), this applies to both predictor attributes and the class attribute.

 

If you have both, a nominal predictor attribute and a nominal class attribute, both attributes are binarized and a weighted average (across the values of the nominal predictor attribute) of weighted averages (each being a weighted average across class values) is calculated.

 

So the process is quite different from what happens in CFS.

 

Cheers,

Eibe

 

From: [hidden email]
Sent: Tuesday, 4 December 2018 10:03 AM
To: [hidden email]
Subject: [Wekalist] CorrelationAttributeEval

 


Hello everyone,

I'm using the CorrelationAttributeEval ranking technique in a classification problem. The dataset I'm working on has numerical and nominal attributes in addition to the class variable(nominal).

To my knowledge CorrelationAttributeEval is based on Pearson's correlation coefficient which only works with continuous variables.

I want to understand how does CorrelationAttributeEval work in this case. Are the numeric features discretized and the symmetrical uncertainty measure used to calculate the correlation between the attributes and the class variable as it is the case for CFS?

 

 


_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
To subscribe, unsubscribe, etc., visit 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
|

Re: CorrelationAttributeEval

hudabenhar
I see. Thank you for your help Eibe!

Le mar. 4 déc. 2018 à 01:07, Eibe Frank <[hidden email]> a écrit :

There is some relevant information to be found under “More”:

 

“Nominal attributes are considered on a value by value basis by treating each value as an indicator. An overall correlation for a nominal attribute is arrived at via a weighted average.”

 

Looking at the code (https://svn.cms.waikato.ac.nz/svn/weka/trunk/weka/src/main/java/weka/attributeSelection/CorrelationAttributeEval.java), this applies to both predictor attributes and the class attribute.

 

If you have both, a nominal predictor attribute and a nominal class attribute, both attributes are binarized and a weighted average (across the values of the nominal predictor attribute) of weighted averages (each being a weighted average across class values) is calculated.

 

So the process is quite different from what happens in CFS.

 

Cheers,

Eibe

 

From: [hidden email]
Sent: Tuesday, 4 December 2018 10:03 AM
To: [hidden email]
Subject: [Wekalist] CorrelationAttributeEval

 


Hello everyone,

I'm using the CorrelationAttributeEval ranking technique in a classification problem. The dataset I'm working on has numerical and nominal attributes in addition to the class variable(nominal).

To my knowledge CorrelationAttributeEval is based on Pearson's correlation coefficient which only works with continuous variables.

I want to understand how does CorrelationAttributeEval work in this case. Are the numeric features discretized and the symmetrical uncertainty measure used to calculate the correlation between the attributes and the class variable as it is the case for CFS?

 

 

_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html


--
Houda Benhar
PhD Candidate at Software Project Management Team
National School of Computer Science and Systems Analysis- Rabat, Morocco
+34 6 54 77 38 59 / +212 6 33 42 96 47

_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html