Tensor Utilities

xtensors.permute(X: TensorLike, axes: AxesPermutation) → XTensor

Permute the axes of X

Parameters:

  • X – target tensor

  • axes – An AxesPermutation (list of integers) object that specifies how the axes should be permuted.

xtensors.newdims(X: TensorLike, *, dims: Dims | None = None, coords: Coords | None = None, position: Literal['left', 'right'] = 'left') → XTensor

Pad singleton dimensions to the given tensor with the specified dimension names and coordinates.

Parameters:

  • X – target tensor

  • dims – (optional, xtensors.Dims) a list of strings specifying the new dimension names

  • coords – (optional, xtensors.Coords) a list of arrays specifying the new coordinates

  • position – left or right, where the new dimensions are padded

xtensors.align(X: TensorLike, Y: TensorLike) → Tuple[XTensor, XTensor]

Pad singleton unnamed dimensions to (at most) one of the two tensors so that the returned tensors have the same number of dimnesions.

xtensors.stack(x: Sequence[XTensor], newdim: str | None = None, position: Literal['left', 'right'] = 'left') → XTensor

Stack a sequence of tensors along a new axis

Parameters:

  • x – sequence of tensors

  • newdim – the name of the new dimension

  • position – left or right, where the new dimension is padded

xtensors.index(X: XTensor, *indices: Tuple[DimLike, int], ignore_if_absent: bool = False) → XTensor

Index a tensor on multiple dimensions

Parameters:

  • X – target tensor

  • indices – a series of (dim, index) pairs

  • ignore_if_absent – if True, dimension names that are specified but not found in X will be ignored without raising errors.

xtensors.to_xtensor(x: TensorLike) → XTensor

Coerce a xtensor.TensorLike object to XTensor

xtensors.dimslast(X: XTensor, dims: Sequence[str]) → XTensor

Move named dimensions to the right.

Parameters:

  • X – target tensor

  • dims – a sequence of dimension names

Returns:

a new xtensors.XTensor with dims moved to the right.

xtensors.dimsfirst(X: XTensor, dims: Sequence[str]) → XTensor

Move named dimensions to the left.

Parameters:

  • X – target tensor

  • dims – a sequence of dimension names

Returns:

a new xtensors.XTensor with dims moved to the left

xtensors.flatten(X: TensorLike, /, dims: DimsLike, dim_out: str | None, *, position: Literal['left', 'right'] = 'right') → XTensor

Flatten a tensor

Parameters:

  • X – target tensor

  • dims – dimensions to be flattened

  • dim_out – name of the new dimension

  • position – left or right, where the new dimension is placed

xtensors.name_dim_if_absent(X: TensorLike, /, axis: int, dim: str, *, force: bool = False) → XTensor

Ensure that X has a named dimension called dim. If not already, the dimension at axis will be named so.

Parameters:

  • X – target tensor

  • axis – the axis to be renamed

  • dim – axis name

  • force – if True, original axis name will be overwritten