kennard_stone.utils package

kennard_stone.utils package#

Distance computation and helpers.

kennard_stone.utils.is_installed(package_name: str) → bool#

Check if the package is installed.

Parameters:

package_name (str) – package name like sklearn

Returns:

if installed, True

Return type:

bool

kennard_stone.utils.pairwise_distances(X: ArrayLike, Y: ArrayLike | None = None, metric: kennard_stone.utils._type_alias.MetricLike = 'euclidean', n_jobs: int | None = None, ensure_all_finite: bool = True, force_all_finite: bool | None = None, device: kennard_stone.utils._type_alias.Device = 'cpu', verbose: int = 1, **kwargs) → ndarray#

Compute the pairwise distance matrix between two point sets.

Wraps scikit-learn’s pairwise_distances and PyTorch’s torch.cdist and chooses an implementation depending on the environment and device.

Parameters:
  • X (array-like of shape (n_samples_1, n_features)) – First set of points.

  • Y (array-like of shape (n_samples_2, n_features), default=None) – Second set of points. If None, distances within X are computed.

  • metric (str or callable, default="euclidean") – Distance metric. If a string, it must be accepted by scikit-learn. With PyTorch, you can use ‘manhattan’, ‘euclidean’, ‘chebyshev’, or ‘minkowski’.

  • n_jobs (int, default=None) – Number of parallel jobs (effective with scikit-learn only).

  • ensure_all_finite (bool, default=True) – Whether to raise on inf/NaN.

  • force_all_finite (bool | None, default=None) – Deprecated alias of ensure_all_finite. If provided, overrides it.

  • device ({"cpu", "cuda", "mps"} or torch.device or str, default="cpu") – Device to use for computation.

Returns:

Distance matrix.

Return type:

ndarray of shape (n_samples_1, n_samples_2)