cca_zoo.preprocessing¶

Per-view sklearn preprocessing adapters for multiview pipelines.

PerViewTransformer ¶

PerViewTransformer(
    transformer: BaseEstimator | list[BaseEstimator],
)

Bases: BaseEstimator

Apply an sklearn transformer independently to each view.

Wraps any sklearn transformer (StandardScaler, SimpleImputer, PCA, KernelCenterer, ...) so it can be used as a preprocessing step ahead of a cca_zoo multiview estimator: a fresh clone of transformer is fit on each view separately, so features -- and fitted state such as a scaler's mean or an imputer's fill value -- are never shared across views.

fit/transform take and return a list[ArrayLike] of views, the same convention every cca_zoo estimator uses, so :class:PerViewTransformer composes directly with sklearn.pipeline.Pipeline and a cca_zoo multiview estimator as its final step -- no dedicated multiview Pipeline class is needed.

Parameters:

Name	Type	Description	Default
`transformer`	`BaseEstimator \| list[BaseEstimator]`	An sklearn transformer instance, cloned once per view (e.g. `StandardScaler()`), or a list with one transformer instance per view for heterogeneous preprocessing (e.g. `[StandardScaler(), SimpleImputer()]` when only one view has missing values).	required

Examples:

>>> import numpy as np
>>> from sklearn.pipeline import Pipeline
>>> from sklearn.preprocessing import StandardScaler
>>> from sklearn.decomposition import PCA
>>> from cca_zoo.linear import CCA
>>> from cca_zoo.preprocessing import PerViewTransformer
>>> rng = np.random.default_rng(0)
>>> X1, X2 = rng.standard_normal((50, 20)), rng.standard_normal((50, 15))
>>> pipe = Pipeline([
...     ("scale", PerViewTransformer(StandardScaler())),
...     ("pca", PerViewTransformer(PCA(n_components=5))),
...     ("cca", CCA(latent_dimensions=2)),
... ])
>>> scores = pipe.fit_transform([X1, X2])

Source code in cca_zoo/preprocessing/_view_transformer.py

def __init__(self, transformer: BaseEstimator | list[BaseEstimator]) -> None:
    self.transformer = transformer

fit ¶

fit(
    views: list[ArrayLike], y: None = None
) -> PerViewTransformer

Fit an independent clone of transformer on each view.

Parameters:

Name	Type	Description	Default
`views`	`list[ArrayLike]`	List of arrays, each of shape (n_samples, n_features_i).	required
`y`	`None`	Ignored. Present for scikit-learn API compatibility.	`None`

Returns:

Name	Type	Description
`self`	`PerViewTransformer`	Fitted transformer.

Source code in cca_zoo/preprocessing/_view_transformer.py

def fit(self, views: list[ArrayLike], y: None = None) -> PerViewTransformer:
    """Fit an independent clone of ``transformer`` on each view.

    Args:
        views: List of arrays, each of shape (n_samples, n_features_i).
        y: Ignored. Present for scikit-learn API compatibility.

    Returns:
        self: Fitted transformer.
    """
    validated = validate_views(views, ensure_all_finite=False)
    self.transformers_: list[BaseEstimator] = [
        clone(t).fit(v)
        for t, v in zip(self._transformers_for(len(validated)), validated)
    ]
    return self

transform ¶

transform(views: list[ArrayLike]) -> list[np.ndarray]

Transform each view with its own fitted transformer.

Parameters:

Name	Type	Description	Default
`views`	`list[ArrayLike]`	List of arrays, each of shape (n_samples, n_features_i).	required

Returns:

Type	Description
`list[ndarray]`	List of transformed views, one per fitted transformer.

Raises:

Type	Description
`NotFittedError`	If `fit` has not been called.

Source code in cca_zoo/preprocessing/_view_transformer.py

def transform(self, views: list[ArrayLike]) -> list[np.ndarray]:
    """Transform each view with its own fitted transformer.

    Args:
        views: List of arrays, each of shape (n_samples, n_features_i).

    Returns:
        List of transformed views, one per fitted transformer.

    Raises:
        sklearn.exceptions.NotFittedError: If ``fit`` has not been called.
    """
    check_is_fitted(self)
    validated = validate_views(
        views, min_views=len(self.transformers_), ensure_all_finite=False
    )
    return [t.transform(v) for t, v in zip(self.transformers_, validated)]

fit_transform ¶

fit_transform(
    views: list[ArrayLike], y: None = None
) -> list[np.ndarray]

Fit and then transform the training data.

Parameters:

Name	Type	Description	Default
`views`	`list[ArrayLike]`	List of arrays, each of shape (n_samples, n_features_i).	required
`y`	`None`	Ignored.	`None`

Returns:

Type	Description
`list[ndarray]`	List of transformed views, one per fitted transformer.

Source code in cca_zoo/preprocessing/_view_transformer.py

def fit_transform(self, views: list[ArrayLike], y: None = None) -> list[np.ndarray]:
    """Fit and then transform the training data.

    Args:
        views: List of arrays, each of shape (n_samples, n_features_i).
        y: Ignored.

    Returns:
        List of transformed views, one per fitted transformer.
    """
    return self.fit(views, y).transform(views)

inverse_transform ¶

inverse_transform(
    views: list[ArrayLike],
) -> list[np.ndarray]

Invert each view's transform via its own fitted transformer.

Parameters:

Name	Type	Description	Default
`views`	`list[ArrayLike]`	List of arrays, each of shape (n_samples, n_features_i) in the transformed space -- typically the output of :meth:`transform`.	required

Returns:

Type	Description
`list[ndarray]`	List of views mapped back to their original feature space.

Raises:

Type	Description
`NotFittedError`	If `fit` has not been called.
`AttributeError`	If a view's transformer has no `inverse_transform`.

Source code in cca_zoo/preprocessing/_view_transformer.py

def inverse_transform(self, views: list[ArrayLike]) -> list[np.ndarray]:
    """Invert each view's transform via its own fitted transformer.

    Args:
        views: List of arrays, each of shape (n_samples, n_features_i)
            in the *transformed* space -- typically the output of
            :meth:`transform`.

    Returns:
        List of views mapped back to their original feature space.

    Raises:
        sklearn.exceptions.NotFittedError: If ``fit`` has not been called.
        AttributeError: If a view's transformer has no ``inverse_transform``.
    """
    check_is_fitted(self)
    validated = validate_views(
        views, min_views=len(self.transformers_), ensure_all_finite=False
    )
    return [t.inverse_transform(v) for t, v in zip(self.transformers_, validated)]

__sklearn_tags__ ¶

__sklearn_tags__() -> Tags

Return sklearn tags, corrected for this class's non-standard fit.

Like :class:~cca_zoo._base.BaseModel, fit/transform take a list of per-view arrays, not a single 2-D X, so sklearn's own input validation and common estimator checks don't apply.

Source code in cca_zoo/preprocessing/_view_transformer.py

def __sklearn_tags__(self) -> Tags:
    """Return sklearn tags, corrected for this class's non-standard ``fit``.

    Like :class:`~cca_zoo._base.BaseModel`, ``fit``/``transform`` take a
    *list* of per-view arrays, not a single 2-D ``X``, so sklearn's own
    input validation and common estimator checks don't apply.
    """
    tags = super().__sklearn_tags__()
    tags.no_validation = True
    tags.input_tags.two_d_array = False
    tags._skip_test = True
    return tags