index is included in memory usage by default

[max-sixty]

...and sys.getsizeof returns correct value. Closes #11597

Is this the best implementation for a common method across classes?

[jreback]

you can put this in PandasObject I think

[jreback]

test this on series/Index/Categorical as well. (for Index put a test in test_index on the Base which will run for all indexes).

[max-sixty]

I added onto the existing tests in IndexOps - lmk if that's not OK

[max-sixty]

@jreback Python 2.6 error with delta argument on assertAlmostEqual - should I skip the test on 2.6 or find an alternate way of doing this?

xref #7718

[jreback]

we don't use this

use tm.assert_almost_equal

[jreback]

I thought we disabled the test routines that one shouldn't use - in tm.TestCase
but maybe that's an open issue (or if not can you create one)

we are dropping 2.6 soon - but still like for there to be 1 way to do testing

[max-sixty]

OK - although the functionality is materially worse than the nosetests one (actually kinda peculiar with the bool argument for 3 vs 5):

check_less_precise : bool, default False
Specify comparison precision.
5 digits (False) or 3 digits (True) after decimal points are compared.

[max-sixty]

@jreback assertLess isn't supported either on 2.6?
But we're green

[jreback]

actually I think this should call with deep=True.

[max-sixty]

I don't have enough context to know whether deep should be True or False by default, but I can see the logic behind the default for .memory_usage should being the same as that for sys.getsizeof, given the similar use cases.

Or why are the use cases different? Because a user calling from the system wants accuracy vs. speed but a user calling from pandas wants speed vs. accuracy?

[jreback]

using __sizeof__ should have deep=True which gives the 'best' report of memory used. This is not the default for .memory_usage() because its expensive to compute (potentially). see #11595 where it can be somewhat slow (though the cython impl helps).

[jreback]

e.g. if someone is calling sys.getsizeof(df) then I think its appropriate to give the most accurate (if maybe somewhat non-performant) answer.

[max-sixty]

OK, if it's a material performance drag and a different enough set of use cases, sobeit

[max-sixty]

@jreback updated & green

[jreback]

nothing should raise a TypeError, what does? rather than catching this error, need to fix the underlying.