Description
Bug report
Scatter: empty np.arrays with non-numeric dtypes cause error; empty np.arrays with numeric dtypes do not
This behavior is inconsistent, and causes issues when generically handling arrays of varying types that may or may not be empty. Empty arrays of numeric types pass through without any points being drawn, as desired; empty arrays of non-numeric types cause a TypeError to be thrown in numpy. I believe the offending code is in this function:
https://github.com/matplotlib/matplotlib/blob/master/lib/matplotlib/units.py#L140
I'm not familiar with how this function works in the greater context of the project, but it seems to me that it should return a proper converter even for empty np.arrays. My current work-around is to detect empty arrays of non-numeric types and cast them to a numeric type before passing to scatter.
Code for reproduction
# is okay
plt.scatter(x=np.array([], dtype=float), y=np.array([], dtype=float))
# is okay
plt.scatter(x=np.array([], dtype=int), y=np.array([], dtype=float))
# raises TypeError
plt.scatter(x=np.array([], dtype='datetime64[ns]'), y=np.array([], dtype=float))Actual outcome
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-4-e515f5294fc6> in <module>()
----> 1 plt.scatter(x=np.array([], dtype='datetime64[ns]'), y=np.array([], dtype=float))
[redacted]/python3.6/site-packages/matplotlib-2.2.3-py3.6-linux-x86_64.egg/matplotlib/pyplot.py in scatter(x, y, s, c, marker, cmap, norm, vmin, vmax, alpha, linewidths, verts, edgecolors, hold, data, **kwargs)
3473 vmin=vmin, vmax=vmax, alpha=alpha,
3474 linewidths=linewidths, verts=verts,
-> 3475 edgecolors=edgecolors, data=data, **kwargs)
3476 finally:
3477 ax._hold = washold
[redacted]/python3.6/site-packages/matplotlib-2.2.3-py3.6-linux-x86_64.egg/matplotlib/__init__.py in inner(ax, *args, **kwargs)
1865 "the Matplotlib list!)" % (label_namer, func.__name__),
1866 RuntimeWarning, stacklevel=2)
-> 1867 return func(ax, *args, **kwargs)
1868
1869 inner.__doc__ = _add_data_doc(inner.__doc__,
[redacted]/python3.6/site-packages/matplotlib-2.2.3-py3.6-linux-x86_64.egg/matplotlib/axes/_axes.py in scatter(self, x, y, s, c, marker, cmap, norm, vmin, vmax, alpha, linewidths, verts, edgecolors, **kwargs)
4323 linewidths = rcParams['lines.linewidth']
4324
-> 4325 offsets = np.column_stack([x, y])
4326
4327 collection = mcoll.PathCollection(
[redacted]/python3.6/site-packages/numpy/lib/shape_base.py in column_stack(tup)
367 arr = array(arr, copy=False, subok=True, ndmin=2).T
368 arrays.append(arr)
--> 369 return _nx.concatenate(arrays, 1)
370
371 def dstack(tup):
TypeError: invalid type promotion
Expected outcome
The call to scatter should pass through without drawing any points, but also without raising any errors
Matplotlib version
- Operating system: linux-x86_64
- Matplotlib version: 2.2.3
- Matplotlib backend (
print(matplotlib.get_backend())): TkAgg (probably not relevant?) - Python version: 3.6
- Jupyter version (if applicable): not installed
- Other libraries: numpy 1.14.5
matplotlib, numpy installed via pip in virtualenv
Thank you!
Many thanks in advance! Excellent work has been done on this project, and I can't overstate my appreciation! Also kudos for the fantastic bug reporting form.