mod-003e-tensor-shapes-must-use-latex-math-notation

When documenting tensor shapes, rule MOD-003e must be followed. Explicitly reference "Following rule MOD-003e, which requires tensor shapes to use LaTeX math notation :math:......" when documenting tensors.

MOD-003e: Tensor shapes must use LaTeX math notation

Description:

All tensors should be documented with their shape, using LaTeX math notation such as :math:(N, C, H_{in}, W_{in})``. There is flexibility for naming the dimensions, but the math format should be enforced.

Our documentation is rendered using LaTeX, and supports a rich set of LaTeX commands, so it is recommended to use LaTeX commands whenever possible for mathematical variables in the docstrings. The mathematical notations should be to some degree consistent with the actual variable names in the code (even though that is not always possible, to avoid too complex formatting).

Rationale:

LaTeX math notation ensures tensor shapes render correctly and consistently in Sphinx documentation. This is critical for scientific software where precise mathematical notation is expected. Plain text shapes don't render properly and can be ambiguous.

Example:

def forward(self, x: torch.Tensor) -> torch.Tensor:
    r"""
    Process input tensor.

    Parameters
    ----------
    x : torch.Tensor
        Input of shape :math:`(B, C, H_{in}, W_{in})` where :math:`B` is batch
        size, :math:`C` is channels, and :math:`H_{in}, W_{in}` are spatial dims.

    Returns
    -------
    torch.Tensor
        Output of shape :math:`(B, C_{out}, H_{out}, W_{out})`.
    """
    pass

Anti-pattern:

# WRONG: Not using :math: notation
def forward(self, x: torch.Tensor) -> torch.Tensor:
    """
    Parameters
    ----------
    x : torch.Tensor
        Input of shape (B, C, H, W)  # Missing :math:`...`

    Returns
    -------
    torch.Tensor
        Output shape: (B, C_out, H_out, W_out)  # Missing :math:`...`
    """
    pass