ENH: Pairwise method or symmetric keyword in DataFrame.corr

[maurosilber]

Is your feature request related to a problem?

I wish I could use pandas to do pairwise computations.

Describe the solution you'd like

The DataFrame.corr method accepts a callable to compute a user-specified correlation function. Its docstring says:

callable: callable with input two 1d ndarrays returning a float. Note that the returned matrix from corr have 1 along the diagonals and will be symmetric of the callable's behavior.

I'd like to compute pairwise calculations which are not symmetric. I can think of two possible solutions:

1. Add a symmetric keyword-only argument, changing the signature to:

def corr(
        self,
        method: str | Callable[[np.ndarray, np.ndarray], float] = "pearson",
        min_periods: int = 1,
        *,
        symmetric=True,
    ) -> DataFrame:

It keeps the current behaviour by default, but allows asymmetric callables to be computed. It could also change the behaviour when computing the diagonals, which are now set to 1, independently of the function provided.

2. Add a new pairwise method.

API breaking implications

The first option shouldn't change the API, as it is adding a keyword-only argument which keeps current behaviour as default.

The second option might not be backwards-compatible with code where columns are named pairwise and called via attribute access.

Additional context

A similar issue was raised here: #25726

[jreback]

that issue see discussions in the PR: #43569

[mroeschke]

Thanks for the suggestion, but given the discussion in #43569, agreed that expanding the API of the corr function does not have much support. Closing