How to set nominal attributes for VDFT

[vinodkraman]

Should I pass in a list nom_att with the indices corresponding to the columns of my dataset that are nominal attributes?

[smastelini]

Hi @vinodkraman, non-numerical features are automatically treated as categorical. The nominal_attributes parameter has to be set when the values of the nominal features are numeric but should be treated as nominal.

[smastelini]

If your categorical features are numeric (e.g., integers), you should pass the feature (column) names to nominal_attributes rather than their indices.

[vinodkraman]

Right, however what if I don't have a column names? Concretely, consider the input x = {0:4, 1:2, 2:4, 3:2} where attributes 0, 1, 2, and 3 are are categorical. Would I need to specify nominal_attributes = [0, 1, 2, 3] so that they match the keys of dictionary?

[smastelini]

Yes, exactly.

[vinodkraman]

Awesome, thanks!