Atsushi Sakai

I'm not sure this is related to the lanczos interpolation, [`signal.windows.lanczos`](http://scipy.github.io/devdocs/reference/generated/scipy.signal.windows.lanczos.html?highlight=lanczos#scipy.signal.windows.lanczos) will be added from v1.10.0 by #16533

Ah... I think my PR #16128 looks very bad. I don't know why I used `max` or ` min` for extrapolation.... https://github.com/scipy/scipy/blob/649932fa64747ef585eabe75cf9a39e3799e3eab/scipy/interpolate/_interpolate.py#L533 https://github.com/scipy/scipy/blob/649932fa64747ef585eabe75cf9a39e3799e3eab/scipy/interpolate/_interpolate.py#L542 I think these should be the right...

Thank you. Your PR is welcome.

PR is welcome

Yeah. That is great. I just reviewed code readability and ran my static analyzer, if the mathematical collectiveness needs to be reviewed, maybe we need other specialists to review it.

Give me a week or so to test the new implementation using my data collections. Then I will merge it.

@mdhaber Sorry, I've tried compiling this PR branch on my MacBook environment several times, but the build for scipy is not passing. It seems to be the same issue related...

I believe this PR has been reviewed for a sufficient amount of time, so I will merge it. Thank you for the fantastic performance improvement!! @mdhaber

The reason for this confusion is that [`curve_fit`](https://scipy.github.io/devdocs/reference/generated/scipy.optimize.curve_fit.html#scipy.optimize.curve_fit) uses the variable name `p0` for the initial guess, while the functions `least_sq` and [`least_squares`](https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.least_squares.html), which are read internally by `curve_fit`, use...