TST: numpy 1.9 / matplotlib 1.4.0 on 3.4-64 bit windows

[jreback]

These passed on numpy 1.8/matplotlib 1.3.1 on 34-64 windows
just updated and now failing

======================================================================
ERROR: test_boxplot_empty_column (pandas.tests.test_graphics.TestDataFramePlots)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win-amd64-3.4\pandas\tests\test_graphics.py", line 1980, in test_boxplot_empty_column
    _check_plot_works(df.boxplot, return_type='axes')
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win-amd64-3.4\pandas\tests\test_graphics.py", line 3349, in _check_plot_works
    ret = f(*args, **kwargs)
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win-amd64-3.4\pandas\core\frame.py", line 4921, in boxplot
    **kwds)
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win-amd64-3.4\pandas\tools\plotting.py", line 2634, in boxplot
    result = plot_group(columns, data.values.T, ax)
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win-amd64-3.4\pandas\tools\plotting.py", line 2585, in plot_group
    bp = ax.boxplot(values, **kwds)
  File "C:\Python34-64\lib\site-packages\matplotlib\axes\_axes.py", line 3045, in boxplot
    labels=labels)
  File "C:\Python34-64\lib\site-packages\matplotlib\cbook.py", line 1965, in boxplot_stats
    q1, med, q3 = np.percentile(x, [25, 50, 75])
  File "C:\Python34-64\lib\site-packages\numpy\lib\function_base.py", line 3042, in percentile
    interpolation=interpolation)
  File "C:\Python34-64\lib\site-packages\numpy\lib\function_base.py", line 2791, in _ureduce
    r = func(a, **kwargs)
  File "C:\Python34-64\lib\site-packages\numpy\lib\function_base.py", line 3131, in _percentile
    x1 = take(ap, indices_below, axis=axis) * weights_below
  File "C:\Python34-64\lib\site-packages\numpy\core\fromnumeric.py", line 121, in take
    return take(indices, axis, out, mode)
IndexError: cannot do a non-empty take from an empty axes.

======================================================================
FAIL: test_boxplot (pandas.tests.test_graphics.TestDataFramePlots)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win-amd64-3.4\pandas\tests\test_graphics.py", line 1794, in test_boxplot
    self.assertEqual(len(ax.lines), 8 * len(numeric_cols))
AssertionError: 21 != 24

======================================================================
FAIL: test_boxplot_vertical (pandas.tests.test_graphics.TestDataFramePlots)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win-amd64-3.4\pandas\tests\test_graphics.py", line 1828, in test_boxplot_vertical
    self.assertEqual(len(ax.lines), 8 * len(numeric_cols))
AssertionError: 21 != 24

----------------------------------------------------------------------
Ran 7882 tests in 724.270s

FAILED (SKIP=267, errors=1, failures=2)

[jreback]

@TomAugspurger
cc @sinhrks

[TomAugspurger]

Will do tomorrow. See also #8296 so this may not be a windows specific thing. #8296 just had 2/3 failures, but I recall seeing that third failure another time.

[jreback]

@TomAugspurger thanks, yes these fail on linux too (on travis we are not testing this ATM)

[tacaswell]

I would check the boxplot failures with mpl-master. We have a couple of (now fixed) bugs in boxplot.

cc @phobson

[TomAugspurger]

OK there were two changes that affected our tests (I'm running mpl and pandas off the latest mater).

1. Handling of empty arrays. In test_graphics:TestDataFramePlots.test_boxplot_empty_column
   One of the columns in the DataFrame is all NaNs. I'll try to come up with a self contained example, but for now (EDIT: Added a cleaner example in comment below):

Here's the matplotlib portion of the stack trace:

  /Users/tom/Envs/py3/lib/python3.4/site-packages/matplotlib-1.5.x-py3.4-macosx-10.9-x86_64.egg/matplotlib/axes/_axes.py(3045)boxplot()
   3043         """
   3044         bxpstats = cbook.boxplot_stats(x, whis=whis, bootstrap=bootstrap,
-> 3045                                        labels=labels)
   3046         if flierprops is None:
   3047             flierprops = dict(sym=sym)

  /Users/tom/Envs/py3/lib/python3.4/site-packages/matplotlib-1.5.x-py3.4-macosx-10.9-x86_64.egg/matplotlib/cbook.py(1966)boxplot_stats()
   1964 
   1965         # medians and quartiles
-> 1966         q1, med, q3 = np.percentile(x, [25, 50, 75])
   1967 
   1968         # interquartile range

  /Users/tom/Envs/py3/lib/python3.4/site-packages/numpy/lib/function_base.py(2820)percentile()
   2818         axis = 0
   2819 
-> 2820     return _compute_qth_percentile(sorted, q, axis, out)
   2821 
   2822 # handle sequence of q's without calling sort multiple times

  /Users/tom/Envs/py3/lib/python3.4/site-packages/numpy/lib/function_base.py(2826)_compute_qth_percentile()
   2824     if not isscalar(q):
   2825         p = [_compute_qth_percentile(sorted, qi, axis, None)
-> 2826              for qi in q]
   2827 
   2828         if out is not None:

> /Users/tom/Envs/py3/lib/python3.4/site-packages/numpy/lib/function_base.py(2826)<listcomp>()
   2824     if not isscalar(q):
   2825         p = [_compute_qth_percentile(sorted, qi, axis, None)
-> 2826              for qi in q]
   2827 
   2828         if out is not None:

  /Users/tom/Envs/py3/lib/python3.4/site-packages/numpy/lib/function_base.py(2856)_compute_qth_percentile()
   2854     # Use add.reduce in both cases to coerce data type as well as
   2855     #   check and use out array.
-> 2856     return add.reduce(sorted[indexer]*weights, axis=axis, out=out)/sumval

ValueError: operands could not be broadcast together with shapes (0,) (2,) 

Here are the values:

ipdb> sorted[indexer]
array([], dtype=float64)
ipdb> weights
array([ 1.25, -0.25])
ipdb> indexer
[slice(0, 2, None)]
ipdb> sorted[indexer] * weights
*** ValueError: operands could not be broadcast together with shapes (0,) (2,)

At first glance this looks like a matplotib regression I think.

---

The scond change: a change in the number of lines per box in the boxplot. Before it was 8, now it's 7.

In [14]: fig, ax = plt.subplots()

In [15]: ax.boxplot(np.random.randn(30))
Out[15]: 
{'means': [],
 'caps': [<matplotlib.lines.Line2D at 0x111239e80>,
  <matplotlib.lines.Line2D at 0x11123e908>],
 'boxes': [<matplotlib.lines.Line2D at 0x111239a20>],
 'fliers': [<matplotlib.lines.Line2D at 0x111816908>],
 'whiskers': [<matplotlib.lines.Line2D at 0x111232208>,
  <matplotlib.lines.Line2D at 0x11122f6d8>],
 'medians': [<matplotlib.lines.Line2D at 0x111816128>]}

This one looks to be intentional, is that correct @phobson, @tacaswell ? Our tests were expecting 8 lines per box. We can just change that to 7.

[TomAugspurger]

Smaller example for the first one:

In [14]: fig, ax = plt.subplots()

In [15]: ax.boxplot([np.array([]), np.array([1, 2, 3, 4, 5])])
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-15-febecc8c1634> in <module>()
----> 1 ax.boxplot([np.array([]), np.array([1, 2, 3, 4, 5])])

/Users/tom/Envs/py3/lib/python3.4/site-packages/matplotlib-1.5.x-py3.4-macosx-10.9-x86_64.egg/matplotlib/axes/_axes.py in boxplot(self, x, notch, sym, vert, whis, positions, widths, patch_artist, bootstrap, usermedians, conf_intervals, meanline, showmeans, showcaps, showbox, showfliers, boxprops, labels, flierprops, medianprops, meanprops, capprops, whiskerprops, manage_xticks)

   3044         bxpstats = cbook.boxplot_stats(x, whis=whis, bootstrap=bootstrap,
-> 3045                                        labels=labels)
   3046         if flierprops is None:
   3047             flierprops = dict(sym=sym)

/Users/tom/Envs/py3/lib/python3.4/site-packages/matplotlib-1.5.x-py3.4-macosx-10.9-x86_64.egg/matplotlib/cbook.py in boxplot_stats(X, whis, bootstrap, labels)
   1964 
   1965         # medians and quartiles
-> 1966         q1, med, q3 = np.percentile(x, [25, 50, 75])
   1967 
   1968         # interquartile range

/Users/tom/Envs/py3/lib/python3.4/site-packages/numpy/lib/function_base.py in percentile(a, q, axis, out, overwrite_input)
   2818         axis = 0
   2819 
-> 2820     return _compute_qth_percentile(sorted, q, axis, out)
   2821 
   2822 # handle sequence of q's without calling sort multiple times

/Users/tom/Envs/py3/lib/python3.4/site-packages/numpy/lib/function_base.py in _compute_qth_percentile(sorted, q, axis, out)
   2824     if not isscalar(q):
   2825         p = [_compute_qth_percentile(sorted, qi, axis, None)
-> 2826              for qi in q]
   2827 
   2828         if out is not None:

/Users/tom/Envs/py3/lib/python3.4/site-packages/numpy/lib/function_base.py in <listcomp>(.0)
   2824     if not isscalar(q):
   2825         p = [_compute_qth_percentile(sorted, qi, axis, None)
-> 2826              for qi in q]
   2827 
   2828         if out is not None:

/Users/tom/Envs/py3/lib/python3.4/site-packages/numpy/lib/function_base.py in _compute_qth_percentile(sorted, q, axis, out)
   2854     # Use add.reduce in both cases to coerce data type as well as
   2855     #   check and use out array.
-> 2856     return add.reduce(sorted[indexer]*weights, axis=axis, out=out)/sumval
   2857 
   2858 def trapz(y, x=None, dx=1.0, axis=-1):

ValueError: operands could not be broadcast together with shapes (0,) (2,) 

[tacaswell]

We have an outstanding issue regarding the fliers so I am not sure which way that one is going to go.

I agree the ValueError one looks like a regression.

[TomAugspurger]

Here's the fliers issue.

@jreback OK with skipping these for now? I'll skip if we detect mpl 1.4 or dev (which is labeled `'1.5.x'). When those are closed I'll come back and remove the skips for 1.5 & possibly change some of the tests depending on how the fliers issue goes.

[jreback]

sure that's fine, just create a reminder issue for next version!