Tensor shape

[lars-reimann]

Is your feature request related to a problem?

Layer.input_size and Layer.output_size currently return int | ModelImageSize. A user must first type check the result to do anything with it.

Desired solution

• Add a class TensorShape with
  • Property dimensionality: int. It should contain the number of indices required to access a scalar in the tensor
  • Method get_size(dimension: int) -> int. It should return the number of elements along the given dimension.
    • Raise if dimension < 0 (0-based indexing) - use _check_bounds.
    • Return 0 if dimension >= dimensionality.
• Make the existing ModelImageSize a subclass of TensorShape.
• Adjust Layer.input_size and Layer.output_size to return a TensorShape instead.

Possible alternatives (optional)

No response

Screenshots (optional)

No response

Additional Context (optional)

No response