src.prepro.prepro_v2

Version 2 preprocessing mixins that compute scaling parameters without applying them.

These mixins are designed to work with the in-graph scaling architecture where scaling is applied within the PyMC model using pm.Data containers, not during preprocessing.

Module Contents

class src.prepro.prepro_v2.MaxAbsScaleTargetV2(*args, **kwargs)

Mixin that computes target scaling parameters without applying them.

This is the v2 version designed for in-graph scaling. It computes the scaling parameters during preprocessing but doesn’t apply them - they are applied later within the PyMC model graph.

preprocess_target(target: pandas.Series) → pandas.Series

Computes target scaling parameters without applying them.

Parameters:

target – The target variable series

Returns:

The original unscaled target (scaling will be applied in-graph)

get_target_transformer() → sklearn.preprocessing.MaxAbsScaler | None

Returns the fitted target transformer for use in inverse transforms.

get_target_scaling_params() → Dict[str, float]

Returns the computed target scaling parameters.

Returns:

Dictionary with ‘scale’ parameter for the target

class src.prepro.prepro_v2.MaxAbsScaleChannelsV2(*args, **kwargs)

Mixin that computes channel scaling parameters without applying them.

This is the v2 version designed for in-graph scaling. It computes the scaling parameters during preprocessing but doesn’t apply them - they are applied later within the PyMC model graph.

preprocess_features(X: pandas.DataFrame, channel_columns: list | None = None) → pandas.DataFrame

Computes channel scaling parameters without applying them.

Parameters:

• X – The feature dataframe

• channel_columns – List of channel column names

Returns:

The original unscaled features (scaling will be applied in-graph)

get_channel_scaling_params() → Dict[str, numpy.ndarray]

Returns the computed channel scaling parameters.

Returns:

Dictionary with ‘scale’ parameter array for channels