Weighted k Nearest Neighbours bug

Hi

 

When using the k Nearest Neighbours node selecting weighted results in class probabilities of 0/NaN or NaN/0 for the two classes. Unweighted sucessfully calculates the probabilities. 

 

Details

Class column: string value of 1.0 or -1.0. 

Descripotrs: 896 int columns containing 0 or 1

 

The weighted probabilities are returnes as NaN for the predicted class and 0 for the other class. 

 

Offending training and validation tables included in the zip file attached.

Cheers

 

Sam