ENH: Set numeric_only=True in aggregations

[WillAyd]

I've noticed that I make this mistake quite often:

>>> df = pd.DataFrame({"a": ["1"] * 3, "b": np.ones(3)})
>>> df.sum()
a    111
b      3
dtype: object

Getting 111 as a result in column a is harmless in this example, but actually quite annoying in most real life use cases where it can produce exceedingly large strings that exhaust memory or tie up the interpreter.

The numeric_only argument gets around this issue:

>>> df.sum(numeric_only=True)
b    3.0
dtype: float64

Though I'm curious if this should really be the default

[jorisvandenbossche]

There should already be some issues about this, eg I remember @jbrockmendel recently raising this as well, and I suspect some older issues as well

[jorisvandenbossche]

The one I remembered is #28900, I think this can be closed as a duplicate?

[WillAyd]

It's a little bit different. That PR is about removing the kwarg whereas this is about changing the default value

The former would encapsulate the latter but we may want to do in steps instead of outright removal

[jorisvandenbossche]

OK, that's right, but I would still propose to use the other issue for that discussion? As in the end it is a single discussion about what to do with this keyword: keep as is, change default, or fully remove. It are not really separate discussions?
So I would put your proposal to change the default (instead of removing the kwarg) in #28900

[WillAyd]

Sure that works just as well