Problem
The current implementation of filter methods based on this work appears to only support binary labels (at least that is my impression based on this line).
Possible solutions
Ideally the full method described in the paper could be implemented and maybe even extended to regression labels where classes are formed by a qcut on the label column.
In lieu of a full implementation, it would be great if the limitation to binary labels could be articulated clearly in the function definition or even better catch label columns that are not binary and error. There is not dedicated to error handling and in particular if a label is multi-class but includes 0 and 1 say [0,0,1,3,4,2,0...] the filter method will appear to work without error unless a particular node has no 0,1 labels in it.
Problem
The current implementation of filter methods based on this work appears to only support binary labels (at least that is my impression based on this line).
Possible solutions
Ideally the full method described in the paper could be implemented and maybe even extended to regression labels where classes are formed by a
qcuton the label column.In lieu of a full implementation, it would be great if the limitation to binary labels could be articulated clearly in the function definition or even better catch label columns that are not binary and error. There is not dedicated to error handling and in particular if a label is multi-class but includes
0and1say[0,0,1,3,4,2,0...]the filter method will appear to work without error unless a particular node has no 0,1 labels in it.