ENH: rank() should return 0 indexed values for consistency

[lcrmorin]

Is your feature request related to a problem?

I am using pandas for a ranking task. The default series.rank() method returns ranks that starts at 1 instead of 0. Thus breaking some unexpected stuff if assuming 0 indexed result, as is common in python.

Describe the solution you'd like

Series.rank() should return 0 indexed values.

API breaking implications

Breaking existing code that rely on 1-indexed ranks

Describe alternatives you've considered

Shifting the output by -1 is the obvious solution, but feels weird.

[mzeitlin11]

Thanks for the request @lcrmorin.

-1 here since this would cause a bunch of churn (deprecation warnings for all users of rank about the changing result is brutal, so would probably have to add another arg instead) and the alternative is an easy way to achieve this behavior.

I'm also not sure a 0-indexed result is better - rank is not necessarily meant to be used as an indexing tool - the default method is average so users often end up with non-whole number values anyway. Th indexing use case sounds more like what argsort would do.

[attack68]

argsort sounds like the way to go here. Although the documentation should probably link to this. Also argsort needs fixing -> #42090