MultivariateTransformer

class pyts.multivariate.transformation.MultivariateTransformer(estimator: _SupportsFitTransform | list[_SupportsFitTransform], flatten: bool = True)

Transformer for multivariate time series.

It provides a convenient class to transform multivariate time series with transformers that can only deal with univariate time series.

Parameters:

estimatorestimator object or list thereof

Transformer. If one estimator is provided, it is cloned and each clone transforms one feature. If a list of estimators is provided, each estimator transforms one feature.

flattenbool (default = True)

Affect shape of transform output. If True, transform returns an array with shape (n_samples, *). If False, the output of transform from each estimator must have the same shape and transform returns an array with shape (n_samples, n_features, *). Ignored if the transformers return sparse matrices.

Attributes:

estimators_list of estimator objects

The collection of fitted transformers.

Examples

>>> from pyts.datasets import load_basic_motions
>>> from pyts.multivariate.transformation import MultivariateTransformer
>>> from pyts.image import GramianAngularField
>>> X, _, _, _ = load_basic_motions(return_X_y=True)
>>> transformer = MultivariateTransformer(GramianAngularField(),
...                                       flatten=False)
>>> X_new = transformer.fit_transform(X)
>>> X_new.shape
(40, 6, 100, 100)

Methods

__init__(estimator[, flatten])
fit(X[, y])	Pass.
fit_transform(X[, y])	Fit to data, then transform it.
get_metadata_routing()	Get metadata routing of this object.
get_params([deep])	Get parameters for this estimator.
set_params(**params)	Set the parameters of this estimator.
transform(X)	Apply transform to each feature.

__init__(estimator: _SupportsFitTransform | list[_SupportsFitTransform], flatten: bool = True) → None

fit(X: ArrayLike, y: ArrayLike | None = None) → Self

Pass.

Parameters:

Xarray-like, shape = (n_samples, n_features, n_timestamps)

Multivariate time series.

yNone or array-like, shape = (n_samples,) (default = None)

Class labels.

Returns:

selfobject

fit_transform(X: ArrayLike, y: ArrayLike | None = None) → NDArray[Any] | csr_matrix

Fit to data, then transform it.

Unlike most transformers, transform may return a sparse matrix (when the wrapped per-feature transformers do), so this method is overridden instead of relying on MultivariateTransformerMixin.fit_transform, whose declared return type is a plain array.

Parameters:

Xarray-like, shape = (n_samples, n_features, n_timestamps)

Multivariate time series.

yNone or array-like, shape = (n_samples,) (default = None)

Class labels.

Returns:

X_newarray, shape = (n_samples, *) or (n_samples, n_features, *)

Transformed time series. May be a sparse matrix instead when the wrapped per-feature transformers return one.

transform(X: ArrayLike) → NDArray[Any] | csr_matrix

Apply transform to each feature.

Parameters:

Xarray-like, shape = (n_samples, n_features, n_timestamps)

Multivariate time series.

Returns:

X_newarray, shape = (n_samples, *) or (n_samples, n_features, *)

Transformed time series.