When I use Auto Model, there are Correlation value, ID-ness value, ect.

User42010User42010 Member, University Professor Posts: 5 University Professor
edited March 2020 in Help

The Attribute value with a color-coded status bubble (red / yellow / green) are shown by the quality bars (C / I / S / M / T). How to calculate the correlation shown by C?

 

First I thought it is Pearson correlation coefficient. But after i tested Pearson correlation coefficient method , I did not get the same result as shown by auto model.

 

On the left side auto model there is weight by correlation, I tested that by Pearson correlation coefficient method and the results were the same as given by auto model.

 

What is the difference between the first correlation the weight by correlation


Answers

  • yyhuangyyhuang Administrator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 364 RM Data Scientist
    Dear prof @User42010,

    Do you mean the data quality measures?



    The correlation is the Pearson Correlation coefficient between the attributes and the label. I tested it with Sonar data, and the AutoModel gives same quality measures as the operators. Please give it a try.

    Best,

    YY

  • User42010User42010 Member, University Professor Posts: 5 University Professor
    Dear yyhuang, 
    Thx for your reply. 
    Actually my results (data quality measures) are the same as yours for this part. But when I run correlation in excel for the same dataset (sonar) I got different results. The results I got when I ran correlation in excel is the same results that Weight by Correlation (on the left bottom side of the Results' window) gives. the results' screen and Weight by Correlation results are attached.
    Notice the correlation of attribute_2 in the screen you sent and the Weight by Correlation in the screen that am attaching are different.

    So now I believe that Correlation is not the same as Weight by Correlation. So, what are the formulas used for each of them?


     

Sign In or Register to comment.