plaid.pipelines.plaid_blocks¶

Custom meta-estimators for applying feature-wise and target-wise transformations.

Includes:

PlaidTransformedTargetRegressor: transforms the target before fitting.
PlaidColumnTransformer: applies transformers to feature subsets like ColumnTransformer.

Attributes¶

Self

Classes¶

`ColumnTransformer`	Custom column-wise transformer for PLAID-style datasets.
`TransformedTargetRegressor`	Meta-estimator that transforms the target before fit and inverses it at predict.

Module Contents¶

Self[source]¶

class ColumnTransformer(plaid_transformers: list[tuple[str, sklearn.base.TransformerMixin | sklearn.pipeline.Pipeline]])[source]¶

Bases: sklearn.compose.ColumnTransformer

Custom column-wise transformer for PLAID-style datasets.

Similar to scikit-learn’s ColumnTransformer, this class applies a list of transformer blocks to subsets of features, defined by their feature identifiers. Additionally, it preserves a set of remainder features that bypass transformation.

Parameters:: plaid_transformers – A list of tuples (name, transformer), where each transformer is a TransformerMixin.

Note

At fit, it is checked that plaid_transformers share no in_features_identifiers and no out_features_identifiers.

plaid_transformers[source]¶

fit(dataset: plaid.Dataset, _y=None) → Self[source]¶

Fits all transformers on their corresponding feature subsets.

Parameters:

dataset – A Dataset object or a list of samples.
y – Ignored. Present for API compatibility.

Returns:

The fitted PlaidColumnTransformer.

Return type:

self

transform(dataset: plaid.Dataset) → plaid.Dataset[source]¶

Applies fitted transformers to feature subsets and merges results.

Parameters:: dataset – A Dataset object or a list of samples.
Returns:: A new Dataset with transformed feature blocks, including untransformed remainder features.
Return type:: Dataset

fit_transform(dataset: plaid.Dataset, y=None) → plaid.Dataset[source]¶

Fits all transformers and returns the combined transformed dataset.

Parameters:

dataset – A Dataset object or a list of samples.
y – Ignored. Present for API compatibility.

Returns:

A new Dataset with transformed features.

Return type:

Dataset

inverse_transform(dataset: plaid.Dataset) → plaid.Dataset[source]¶

Applies fitted inverse transformers to feature subsets and merges results.

Parameters:: dataset – A Dataset object or a list of samples.
Returns:: A new Dataset with inverse transformed feature blocks, including untransformed remainder features.
Return type:: Dataset

class TransformedTargetRegressor(regressor: sklearn.base.RegressorMixin | sklearn.pipeline.Pipeline, transformer: sklearn.base.TransformerMixin | sklearn.pipeline.Pipeline)[source]¶

Bases: sklearn.base.RegressorMixin, sklearn.base.BaseEstimator

Meta-estimator that transforms the target before fit and inverses it at predict.

This regressor is compatible with custom Dataset objects and supports complex targets, including scalars and fields. It wraps a base regressor and a transformer that is responsible for preprocessing the target space.

Parameters:

regressor – A regressor implementing fit and predict, following the scikit-learn API.
transformer – A transformer implementing fit, transform, and inverse_transform. Applied to the dataset before fitting the regressor.

regressor[source]¶

transformer[source]¶

fit(dataset: plaid.Dataset, _y=None) → Self[source]¶

Fits the transformer and the regressor on the transformed dataset.

Parameters:

dataset – A Dataset object or a list of sample dictionaries. Input training data.
y – Ignored. Present for API compatibility.

Returns:

The fitted estimator.

Return type:

self

predict(dataset: plaid.Dataset) → plaid.Dataset[source]¶

Predicts target values using the fitted regressor, then applies the inverse transformation.

Parameters:: dataset – A Dataset object or a list of sample dictionaries. Input data to predict on.
Returns:: A Dataset containing the inverse-transformed predictions.
Return type:: Dataset

score(dataset_X: plaid.Dataset, dataset_y: plaid.Dataset = None) → float[source]¶

Computes a normalized root mean squared error (RMSE) score on the transformed targets.

The score is defined as 1 - avg(relative RMSE) over all target features in the transformer input features identifiers. The error computation depends on the feature type: - For “scalar” features: RMSE normalized by squared reference value. - For “field” features: RMSE normalized by field size and max-norm of the reference.

Parameters:

dataset_X – A Dataset object or a list of samples. Input features used for prediction.
dataset_y – A Dataset object or list, optional. Ground-truth targets. If None, dataset_X is used for both input and reference.

Returns:

A score between -inf and 1. A perfect prediction yields a score of 1.0.

Return type:

float

Raises:

ValueError – If an unknown feature type is encountered.