Currently, for all scikit learn Khiops estimators, if the observables X are in a pandas.DataFrame, then the targets y can either be in a pandas.Series or in a (single-column) pandas.DataFrame.
The goal of this issue is to allow targets y to be array-like, that is, to be in an object which has the __array__ method defined. This includes NumPy arrays.
Questions/Ideas
In order to achieve this, we need to:
make sure type-checking rules allow for the (X: pandas.DataFrame, y: hasattr(y, "__array__") combination
allow (or, rather, complete) support for array-like targets in the PandasTable class
add relevant tests, in test_dataset_class, but also in test_dataset_errors if applicable.
The end result of this issue would be that all in-memory data tables would support array-like targets.
Description
Currently, for all scikit learn Khiops estimators, if the observables
X
are in apandas.DataFrame
, then the targetsy
can either be in apandas.Series
or in a (single-column)pandas.DataFrame
.The goal of this issue is to allow targets
y
to be array-like, that is, to be in an object which has the__array__
method defined. This includes NumPy arrays.Questions/Ideas
X: pandas.DataFrame
,y: hasattr(y, "__array__")
combinationPandasTable
classtest_dataset_class
, but also intest_dataset_errors
if applicable.