jax.lax.linalg.svd#

jax.lax.linalg.svd(x: ArrayLike, *, full_matrices: bool = True, compute_uv: Literal[True], subset_by_index: tuple[int, int] | None = None, algorithm: SvdAlgorithm | None = None) → tuple[Array, Array, Array][source]#
jax.lax.linalg.svd(x: ArrayLike, *, full_matrices: bool = True, compute_uv: Literal[False], subset_by_index: tuple[int, int] | None = None, algorithm: SvdAlgorithm | None = None) → Array
jax.lax.linalg.svd(x: ArrayLike, *, full_matrices: bool = True, compute_uv: bool = True, subset_by_index: tuple[int, int] | None = None, algorithm: SvdAlgorithm | None = None) → Array | tuple[Array, Array, Array]

Singular value decomposition.

Computes the singular value decomposition of an input matrix.

Parameters:

  • x – A batch of matrices with shape [..., m, n].

  • full_matrices – Determines if full or reduced matrices are returned.

  • compute_uv – If True, returns the left singular vectors, the singular values and the adjoint of the right singular vectors. Otherwise, only the singular values are returned.

  • subset_by_index – If None, the entire matrix is returned. Otherwise, returns the singular values and vectors for the given range of indices.

  • algorithm – The SVD algorithm to use. Must be None or a value from SvdAlgorithm.

Returns:

The singular values if compute_uv is False, otherwise returns a triple containing the left singular vectors, the singular values, and the adjoint of the right singular vectors.