NestedTabularMLPipeline

class lightautoml.pipelines.ml.nested_ml_pipe.NestedTabularMLPipeline(ml_algos, force_calc=True, pre_selection=None, features_pipeline=None, post_selection=None, cv=1, n_folds=None, inner_tune=False, refit_tuner=False)

Bases: lightautoml.pipelines.ml.base.MLPipeline

Wrapper for MLPipeline to make it trainable over nested folds.

Limitations:

• Only for TabularMLAlgo

• Nested trained only MLAlgo. FeaturesPipelines and SelectionPipelines are trained as usual.

__init__(ml_algos, force_calc=True, pre_selection=None, features_pipeline=None, post_selection=None, cv=1, n_folds=None, inner_tune=False, refit_tuner=False)

Parameters

• ml_algos (Sequence[Union[TabularMLAlgo, Tuple[TabularMLAlgo, ParamsTuner]]]) – Sequence of MLAlgo’s or Pair - (MlAlgo, ParamsTuner).

• force_calc (Union[bool, Sequence[bool]]) – Flag if single fold of MlAlgo should be calculated anyway.

• pre_selection (Optional[SelectionPipeline]) – Initial feature selection. If None there is no initial selection.

• features_pipeline (Optional[FeaturesPipeline]) – Composition of feature transforms.

• post_selection (Optional[SelectionPipeline]) – Post feature selection. If None there is no post selection.

• cv (int) – Nested folds cv split.

• n_folds (Optional[int]) – Limit of valid iterations from cv.

• inner_tune (bool) – Should we refit tuner each inner cv run or tune ones on outer cv.

• refit_tuner (bool) – Should we refit tuner each inner loop with inner_tune==True.