Rounding to significant figures

[owrior]

Description

Recently I have been working on a project where we needed to round to significant figures and realised there is no way to do this in pandas, numpy or polars. I found this as the best solution.

def signif(x, p):
    x = np.asarray(x)
    x_positive = np.where(np.isfinite(x) & (x != 0), np.abs(x), 10**(p-1))
    mags = 10 ** (p - 1 - np.floor(np.log10(x_positive)))
    return np.round(x * mags) / mags

However it would be nice to have it as part of a package. Would it be possible to include this?

I have implemented this already here #11959 which is about 20% faster than the numpy one here already. However I can imagine there is a better way than using the zip_with maybe?

Thank you!

[orlp]

Perhaps this could be a boolean keyword-only argument for round?

Expr.round(decimals: int = 0, *, significant = False)

If set to True it would round to the specified number of significant digits. Alternatively it could be a dedicated function round_significant. Just bikeshedding.

[owrior]

I was thinking that but then thought it might be distinct enough (given the use with integers to) to have in a separate function. Happy to go with what you would prefer though! In the PR I called it round_sf - though maybe that is not clear enough so happy to refactor to round_significant.

[owrior]

@orlp should I make the name more expressive or anything?

[orlp]

@owrior

I had a discussion with @stinodego and I believe we settled on adding this functionality, but as a separate function named round_sig_figs(digits).

[owrior]

@orlp

Thank you very much for this it will be a great help. Can you let me know if my PR is looking good? I have refactored to this function as requested.