`zetta_utils.tensor_ops`¶

Handling tensor operations. Each opperation must support both np.ndarray and torch.Tensor input types.

`zetta_utils.tensor_ops.convert`¶

zetta_utils.tensor_ops.convert.to_np(data)[source]¶

Convert the given tensor to numpy.ndarray.

Parameters:: data (Union[Tensor, TypeAliasType]) – Input tensor_ops.
Return type:: TypeAliasType
Returns:: Input tensor in numpy.ndarray format.

zetta_utils.tensor_ops.convert.to_torch(data, device=None)[source]¶

Convert the given tensor to torch.Tensor.

Parameters:

data (Union[Tensor, TypeAliasType]) – Input tensor_ops.
device (UnionType[device, str, int, None]) – Device name on which the torch tensor will reside.

Return type:

Tensor

Returns:

Input tensor in torch.Tensor format.

zetta_utils.tensor_ops.convert.astype(data, reference, cast=False)[source]¶

Convert the given tensor to np.ndarray or torch.Tensor depending on the type of reference tensor_ops.

Parameters:

data (Union[Tensor, TypeAliasType]) – Input tensor_ops.
reference (TypeVar(TensorTypeVar, Tensor, TypeAliasType, Union[Tensor, TypeAliasType])) – Reference type tensor_ops.
cast (bool) – If True, cast data to the type of reference. Raises ValueError if that would round integers: magnitude above 2**24 for float32/complex64, above 2**53 for float64/complex128.

Return type:

TypeVar(TensorTypeVar, Tensor, TypeAliasType, Union[Tensor, TypeAliasType])

Returns:

Input tensor converted to the reference type.

zetta_utils.tensor_ops.interpolate(data, *args, targets=None, **kwargs)[source]¶

Wraps a tensor op so it also applies valuewise to a Mapping of tensors.

Parameters:

fn – The wrapped tensor op.
targets (Optional[Container[str]]) – If set, only these Mapping keys are processed; others pass through. Set via with_targets().