ENH: Allow function in Series/DataFrame get syntax

[soaxelbrooke]

Currently, pandas makes clever use of __getitem__ to filter/mask Series and DataFrame. At times, the current abstraction is leaky - for instance, df[18 < df['age'] < 25] would raise an exception, even though cascaded compare symbols in that manner are legal Python syntax.

It would be awesome to also be able to pass in a function to the __getitem__ function that would act as a filter. This would allow users to use all of Python's language features for filtering while still maintaining the brevity intended for these statements.

For example:

df[lambda row: 18 < row['age'] < 25]

This would also allow lengthier filters to be defined outside of the getter square brackets:

def age_filter(row):
    return 18 < row['age'] < 25 if row['country'] == 'USA' else row['age'] > 13
df[age_filter]

Functions as viable filters would also allow for chaining of filter functions without creating multiple intermediary DataFrames or Series, making optimization of cascaded filters much easier.

[jreback]

this is a dupe of #2560. you can do this using query/eval, see docs here

you can do chained relationships as well.

__getitem__ is already hopelessly overloaded. I don't think this add any clarify. And it will be prone to non-vectorized evaluation, losing performance. All of the operations you show above are vectorizable