TST/API: Simplify testing functions API

[sinhrks]

Derived from #10507. Consider better API for testing functions.

1. How to specify comparison tolerance

Current assert functions have check_less_precise and check_exact flags to specify how to compare values:

• check_exact=True: Use assert_equal, comparison using ==.
• check_exact=False and check_less_precise=False: Use assert_almost_equal, comparing 5 digits after decimal points (default)
• check_exact=False and check_less_precise=True: Use assert_almost_equal, comparing 3 digits after decimal points

There can be single kw to specify above behaviors?

2. Make integrated assert function

I think it's nice to have test functions which supports all pandas objects for users who is starting contribution / using pandas as their dependencies. Changing assertEquals and assertAlmostEquals to internally use current assert_index/series/frame_equal is one idea.

3. Remove assert_almost_equal

Remove assert_almost_equal and use class-based validation method. Also add input type validation to numpy_assert_array_equal.

xref #9895.

[kawochen]

I've seen multiple versions of 2 scattered across different test scripts, such as this. They might be good building blocks.

[jorisvandenbossche]

Regarding the How to specify comparison tolerance, I have two problems with the current interface:

• It is a bit clumsy, as you have two keywords that could perfectly fit into one keyword, making it a bit confusing.
• It is also not very flexible, as it only gives you the option to check all, 5 or 3 digits. For example, if you have values of different order in one frame (eg 0.0001 to 100), then the absolute number of digits cannot be used to check this but you need something relative. Therefore, you have to fall back to checking the .values with numpy.

Specifically for that last point, I think it is nice to look at np.testing.assert_allclose (which is now the recommended way in numpy according to the assert_array_almost_equal docstring), which uses both relative and absolute tolerance.

[sinhrks]

Added 3rd item, I think we should replace assert_almost_equal to precise functions as much.

[mroeschke]

I think a lot of these have addressed in a meantime so closing