ENH: sum() should default to numeric_only=True about pandas

[asishm]

You may want to read through #46072

Do you have any data that backs this up?

Perhaps this could be reverted to help out the majority of analysts/data scientists that just want a convenient tool that works as expected by humans and not an idealistic pure python implementation that exists just to satisfy a textbook.

[travisturenne]

You may want to read through #46072

Wouldn't it have made more sense to treat timedelta as numeric rather than appending every string in a dataframe upon invoking sum()?

[rhshadrach]

Thanks for raising this issue!

Many IDEs for Jupyter notebooks crash when .sum() is used, which requires the user to restart their Jupyter notebook instance.

Can you provide an example that causes your notebook to crash?

When users see code:

df[["A", "B"]].sum()

I think it's reasonable for them to expect the result contains two rows indexed by "A" and "B". This would not be the case if we used numeric_only=True as the default: data would be dropped without any indication from the code. On the other hand:

df[["A", "B"]].sum(numeric_only=True)

makes it very clear that columns may be dropped.

In general, I am against the silent dropping of data, and I think that applies here.