CategoryRole

class lightautoml.dataset.roles.CategoryRole(dtype=<class 'object'>, encoding_type='auto', unknown=5, force_input=False, label_encoded=False, ordinal=False)

Bases: lightautoml.dataset.roles.ColumnRole

Category role.

__init__(dtype=<class 'object'>, encoding_type='auto', unknown=5, force_input=False, label_encoded=False, ordinal=False)

Create category role with specific dtype and attrs.

Parameters

• dtype (Union[Callable, str]) – Variable type.

• encoding_type (str) – Encoding type.

• unknown (int) – Cut-off freq to process rare categories as unseen.

• force_input (bool) – Select a feature for training, regardless of the selector results.

Note

Valid encoding_type:

• ‘auto’ - default processing

• ‘int’ - encode with int

• ‘oof’ - out-of-fold target encoding

• ‘freq’ - frequency encoding

• ‘ohe’ - one hot encoding