Add a new split_models_on key as part of the advanced_args to allow training multiple models for each separate value of a given column.
This should allow us to handle timeseries problems with high granularity group bys.
Currently the implementation should only work for when quick_learn is enabled, since the nonconformist model hooks into lightwood directly and makes this complicated.
Add a new
split_models_on
key as part of theadvanced_args
to allow training multiple models for each separate value of a given column.This should allow us to handle timeseries problems with high granularity group bys.
Currently the implementation should only work for when
quick_learn
is enabled, since the nonconformist model hooks into lightwood directly and makes this complicated.