gyoza.utilities package

Submodules

gyoza.utilities.tensors module

gyoza.utilities.tensors.expand_axes(x: Tensor, axes) → Tensor

Expands x with singleton axes.

Parameters:

• x (tensorflow.Tensor) – The tensor to be expanded.

• axes (List[tensorflow.Tensor]) – The axes along which to expand. Their indices are assumed to be valid in the shape of x_new. This means if, e.g. x has two axes then axes may be, e.g. [0,1,3,5,6,7] where axes 2 and 4 are filled in order by x but axes must not be, e.g. [0,1,3,5,6,10] because of the gap between 6 and 10 that would be introduced in the shape of x_new.

Returns:

x_new (tensorflow.Tensor) - The reshaped version of x with singletons along ‘’axes’’.

gyoza.utilities.tensors.flatten_along_axes(x: Tensor, axes: List[int]) → Tensor

Flattens an input x along axes axes.

Parameters:

• x (tensorflow.Tensor) – The input to be flattened. Assumed to have at least as many axes as indicated by axes.

• axes (List[int]) – The axes along which the input shall be flattened.

Returns:

x_new (tensorflow.Tensor) - The reshaped tensor x flattened along axes.

gyoza.utilities.tensors.move_axis(x: Tensor, from_index: int, to_index: int) → Tensor

Moves an axis from from_index to to_index.

Parameters:

• x (tensorflow.Tensor) – A tensor of shape […, k, …] where k is at from_index.

• from_index (int) – The index of the axis before transposition.

• to_index (int) – The index of the axis after transposition.

Returns:

x_new (tensorflow.Tensor): The tensor x transposed such that shape […, k, …] now has k at to_index.

gyoza.utilities.tensors.swop_axes(x: Tensor, from_axis: int, to_axis: int) → Tensor

Swops axes of x.

Parameters:

• x (tensorflow.Tensor) – The input whose axes shall be swopped. Assumed to have at least as many axes as indicated by from_axis and to_axis.

• from_axis (int) – The axes to be swopped with to_axis.

• to_axis (int) – The axes to be swopped with from_axis.

Returns:

x_new (tensorflow.Tensor) - The input which from_axis and to_axis swopped.

Module contents