hi,
in the performance(Classification) operator, correlation is defined as correlation between the label and the prediction, but how is this calculated? with which informaion criteria? I dont understand the results...
I think its standard pearson correlation with by using the integer mapping of the classes.
~Martin
Hi,
Just looked up the Java doc and here is what it says: Computes the empirical corelation coefficient 'r' between label and prediction. For P=prediction, L=label, V=Variance, Cov=Covariance we calculate r by: Cov(L,P) / sqrt(V(L)*V(P))
P=prediction, L=label, V=Variance, Cov=Covariance
Cov(L,P) / sqrt(V(L)*V(P))
So it is the Pearson correlation between the label and the prediction. Judging from a quick check of the source code itself this description is accurate.
Cheers,
Ingo
