Add support to permute a tensor.

[hzhangxyz]

No description provided.

[Copilot]

Pull Request Overview

This PR adds support for permuting tensor dimensions in the ParityTensor class. The implementation handles the complex parity calculations required when indices are reordered, ensuring that the mathematical properties of the parity tensor are preserved.

• Adds a permute method that reorders tensor dimensions according to a given permutation
• Implements parity sign correction logic to handle index swaps
• Preserves all tensor properties (edges, parity, mask) through the permutation

Comments suppressed due to low confidence (1)

parity_tensor/parity_tensor.py:66

• The parameter name before_by_after is unclear. Consider renaming to dims or permutation to match PyTorch's convention and make the purpose more obvious.

    def permute(self, before_by_after: tuple[int, ...]) -> ParityTensor:

[Copilot]

The docstring is incomplete. It should include parameter descriptions, return value description, and example usage for a public API method.

Suggested change

	Permute the indices of the parity tensor.
	Permute the indices of the parity tensor.
	This method rearranges the dimensions of the tensor and its associated metadata
	(edges, parity, and mask) according to the specified permutation.
	Parameters:
	----------
	before_by_after : tuple[int, ...]
	A tuple specifying the new order of the dimensions. Each element in the tuple
	represents the index of the original dimension that should appear in the
	corresponding position in the new order.
	Returns:
	-------
	ParityTensor
	A new `ParityTensor` instance with permuted indices, updated edges, parity,
	and mask.
	Example:
	--------
	>>> tensor = torch.tensor([[1, 2], [3, 4]])
	>>> edges = ((0, 1), (1, 0))
	>>> parity_tensor = ParityTensor(_edges=edges, _tensor=tensor)
	>>> permuted_tensor = parity_tensor.permute((1, 0))
	>>> print(permuted_tensor.tensor)
	tensor([[1, 3],
	[2, 4]])

[Copilot]

The method lacks input validation. It should verify that before_by_after contains valid indices (0 to tensor.dim()-1) and that all indices are unique to prevent runtime errors.