ENH/PERF: copy=True keyword in methods where copy=False can improve perf · Issue #48141 · pandas-dev/pandas (original) (raw)

xref #48117, #47993, #48043, #47934, #47932

We have a number of methods that currently make copies but don't have to, mainly methods that revolve around index/columns while leaving the data untouched. e.g. DataFrame.droplevel. By adding a keyword copy=True to these methods, we give users the choice to avoid making a copy by setting copy=False, thereby improving performance.

@jorisvandenbossche makes a reasonable point #48117 (review) that adding this may interact with potential copy/view changes in upcoming versions.

Besides droplevel and drop, for which PRs are already open, the other methods I'm looking at are reset_index, replace, and fillna (there may be others I haven't found). In all three of those cases, the idea would be to add copy and deprecate inplace (xref #16529).