API: expose public testing functions in pandas.testing (GH9895)

[jorisvandenbossche]

Closes #9895

This just exposes a few of our testing functions publicly in pandas.testing.

• Alternative is to move the full testing module to pandas.testing (the actual code), question is then however how to only expose a subset of it publicly.
• I now only included the 3 most obvious ones (frame, series, index). Are there other ones that people would want to use? (the sparse ones?)
• We have quite a lot of keywords to those assert_* functions. We could op to only expose a subset of those publicly (redefining in pandas.testing instead of importing and only passing subset of the kwargs through)

[jreback]

lgtm. need some whatsnew docs (but can prob add onto the pandas.errors section maybe (rename it though)

[codecov]

Codecov Report

Merging #16003 into master will increase coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master   #16003      +/-   ##
==========================================
+ Coverage   91.02%   91.02%   +<.01%     
==========================================
  Files         145      146       +1     
  Lines       50391    50394       +3     
==========================================
+ Hits        45870    45873       +3     
  Misses       4521     4521

Flag	Coverage Δ
#multiple	88.83% <100%> (ø)	⬆️
#single	40.33% <100%> (ø)	⬆️

Impacted Files	Coverage Δ
pandas/util/testing.py	81.01% <ø> (ø)	⬆️
pandas/testing.py	100% <100%> (ø)
pandas/__init__.py	92.85% <100%> (+0.17%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 2d9451d...bb65d12. Read the comment docs.

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (master@1f812e3). Click here to learn what that means.
The diff coverage is 100%.

@@            Coverage Diff            @@
##             master   #16003   +/-   ##
=========================================
  Coverage          ?      91%           
=========================================
  Files             ?      147           
  Lines             ?    50422           
  Branches          ?        0           
=========================================
  Hits              ?    45887           
  Misses            ?     4535           
  Partials          ?        0

Flag	Coverage Δ
#multiple	88.81% <100%> (?)
#single	40.32% <100%> (?)

Impacted Files	Coverage Δ
pandas/util/testing.py	81.01% <ø> (ø)
pandas/testing.py	100% <100%> (ø)
pandas/__init__.py	92.85% <100%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 1f812e3...1bf8a93. Read the comment docs.

[jorisvandenbossche]

I did a quick search for which pandas.util.testing functionality is used by other projects (apart from the main 3 assert_frame/series/index_equal ones):

• dask:
  • makeDataFrame, makeTimeDataFrame, makeMissingDataframe, makeDateIndex, makeTimedeltaIndex, makeTimeSeries
  • assertRaisesRegexp, assertRaises, assert_produces_warning, assert_numpy_array_equal
• statsmodels:
  • makeDataFrame, makeTimeDataFrame, assert_produces_warning
• zipline:
  • assert_panel_equal (but since panel is deprecated, this is less important)

I am not sure whether we want to commit to keeping some of the make.. functions stable? (and projects can of course keep using them at own risk as it is now)

For the raises and warning ones, since those are not pandas-objects specific test functions, I would also not include them in the official public ones?

[jreback]

xref #15911

I think we will create a pandas.util.benchmarks sub-module that holds things like make* which we can make public, keeping pandas.util.testing private (or dask can reach in, doesn't really matter). I don't think these need to be really public anyhow.

cc @mrocklin

[jreback]

makeDataFrame in camelCase is a really old naming convention anyhow.

[jreback]

pls rebase. lgtm. let's merge and can always amend/add to this.

[jreback]

I think both of these should be moved to the privacy sub-section now.

[jorisvandenbossche]

indeed, will need some general editing

[Rufflewind]

Does this mean pytest is now a dependency of pandas?

I just upgraded pandas through AUR and got this error message:

Traceback (most recent call last):
  File "/usr/lib/python3.6/site-packages/pandas/__init__.py", line 60, in <module>
    import pandas.testing
  File "/usr/lib/python3.6/site-packages/pandas/testing.py", line 7, in <module>
    from pandas.util.testing import (
  File "/usr/lib/python3.6/site-packages/pandas/util/testing.py", line 22, in <module>
    import pytest
ModuleNotFoundError: No module named 'pytest'

I do not have pytest installed. Is it now required?

[jreback]

pytest is not required unless you are actually testing things - this is fixed in the rc

unclear what version you actually have

[Rufflewind]

I had 1b52b12 installed. Looks like it was fixed in 70ae817 . Re-reinstalled Pandas and it seems to import without pytest just fine now. Thanks!