DEPR: Series.compress

[TomAugspurger]

xref #18262

[codecov]

Codecov Report

Merging #21930 into master will increase coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master   #21930      +/-   ##
==========================================
+ Coverage      92%      92%   +<.01%     
==========================================
  Files         170      170              
  Lines       50609    50611       +2     
==========================================
+ Hits        46561    46563       +2     
  Misses       4048     4048

Flag	Coverage Δ
#multiple	90.4% <100%> (ø)	⬆️
#single	42.19% <0%> (-0.01%)	⬇️

Impacted Files	Coverage Δ
pandas/core/series.py	94.11% <100%> (+0.01%)	⬆️

---

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 3e48393...c228d49. Read the comment docs.

[gfyoung]

I prefer what you said in the warning message, which is to use Series[condition] instead (more end-user friendly IMO vs. Series.__getitem__).

[jreback]

comment by @gfyoung but otherwise, merge away.

[jreback]

thanks @TomAugspurger

[jorisvandenbossche]

One problem with this PR is that now np.compress also raises a warning:

In [9]: cond = [True, False, True, False, False]
   ...:   s = Series([1, -1, 5, 8, 7], index=list('abcde'))

In [10]: np.compress(cond, s)
/home/joris/miniconda3/envs/dev/lib/python3.5/site-packages/numpy/core/fromnumeric.py:57: FutureWarning: Series.compress(condition) is deprecated. Use Series[condition] instead.
  return getattr(obj, method)(*args, **kwds)
Out[10]: 
a    1
c    5
Name: foo, dtype: int64

so apparently np.compress will simply call the method on the passed object if that exists.

Is there something to do about that? (I suppose not?)

[jorisvandenbossche]

Additional problem is that once we remove the method, np.compress will also not work anymore.

You get something like (with the method commented out):

In [2]: np.compress(cond, s)
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
~/miniconda3/envs/dev/lib/python3.5/site-packages/numpy/core/fromnumeric.py in _wrapfunc(obj, method, *args, **kwds)
     56     try:
---> 57         return getattr(obj, method)(*args, **kwds)
     58 

~/scipy/pandas/pandas/core/generic.py in __getattr__(self, name)
   4557                 return self[name]
-> 4558             return object.__getattribute__(self, name)
   4559 

AttributeError: 'Series' object has no attribute 'compress'

During handling of the above exception, another exception occurred:

ValueError                                Traceback (most recent call last)
<ipython-input-2-cb6f12fdb9e8> in <module>()
----> 1 np.compress(cond, s)

~/miniconda3/envs/dev/lib/python3.5/site-packages/numpy/core/fromnumeric.py in compress(condition, a, axis, out)
   1668 
   1669     """
-> 1670     return _wrapfunc(a, 'compress', condition, axis=axis, out=out)
   1671 
   1672 

~/miniconda3/envs/dev/lib/python3.5/site-packages/numpy/core/fromnumeric.py in _wrapfunc(obj, method, *args, **kwds)
     65     # a downstream library like 'pandas'.
     66     except (AttributeError, TypeError):
---> 67         return _wrapit(obj, method, *args, **kwds)
     68 
     69 

~/miniconda3/envs/dev/lib/python3.5/site-packages/numpy/core/fromnumeric.py in _wrapit(obj, method, *args, **kwds)
     49         if not isinstance(result, mu.ndarray):
     50             result = asarray(result)
---> 51         result = wrap(result)
     52     return result
     53 

~/scipy/pandas/pandas/core/series.py in __array_wrap__(self, result, context)
    655         """
    656         return self._constructor(result, index=self.index,
--> 657                                  copy=False).__finalize__(self)
    658 
    659     def __array_prepare__(self, result, context=None):

~/scipy/pandas/pandas/core/series.py in __init__(self, data, index, dtype, name, copy, fastpath)
    264                             'Length of passed values is {val}, '
    265                             'index implies {ind}'
--> 266                             .format(val=len(data), ind=len(index)))
    267                 except TypeError:
    268                     pass

ValueError: Length of passed values is 2, index implies 5

which is not a very user friendly error message ..

Of course, if we are fine with np.compress(series) not working anymore, then the fact that it raises a deprecation message (my previous comment) is of course fine ..

[TomAugspurger]

How about we update the warning message to be

Use `series[condition] or np.asarray(series).compress(condition)`