issue_comments

imshow with all NaNs works for me on matplotlib 1.5.2

ups yes, I might have been confused by the warning, but it does work with latest mpl yes. The warning says:

``` In [1]: import numpy as np

In [2]: import matplotlib.pyplot as plt

In [3]: a = np.full((2, 2), np.NaN)

In [4]: plt.imshow(a)
Out[4]: <matplotlib.image.AxesImage at 0x7f6e2a8ba710>

In [5]: plt.show() /home/mowglie/.pyvirtualenvs/py3/lib/python3.5/site-packages/matplotlib/colors.py:897: UserWarning: Warning: converting a masked element to nan. dtype = np.min_scalar_type(value) /home/mowglie/.pyvirtualenvs/py3/lib/python3.5/site-packages/numpy/ma/core.py:748: UserWarning: Warning: converting a masked element to nan. data = np.array(a, copy=False, subok=subok) ```

@fmaussion what version of matplotlib are you using? imshow with all NaNs works for me on matplotlib 1.5.2 (yes, we're stuck in the dark ages). This might be a regression in matplotlib.

I investigated a bit and it seems that in the case of all nan data you would probably have to force add_colorbar to False and do other things for matplotlib to accept your all nan data.

I don't need to do this, but I'd like to for another feature (imshow RGB images) - the question is how! Passing add_colorbar=False to plot.imshow just gives me errors, and I can't find anywhere in the internals where this would be valid either - it always gets passed through to the matplotlib Artist and then an exception. Any ideas?

I'm willing to write the patch, if that helps!

Yes, thanks! The best place to start would be to add a test to Common2dMixin (https://github.com/pydata/xarray/blob/master/xarray/tests/test_plot.py#L560).

I investigated a bit and it seems that in the case of all nan data you would probably have to force add_colorbar to False and do other things for matplotlib to accept your all nan data.

I'd argue that Xarray should handle these cases - it already does for lower dimensions, eg xr.DataArray(np.full(1, np.nan)).plot() works, and this would be really nice for exploratory analysis as well as consistency.

This is sufficiently annoying to my research group that I'm willing to write the patch, if that helps! I also have some ideas for tests using Hypothesis to ferret out some other problems - for example there's a similar failure if plotting an array with size one in some dimension.

Actually I changed my mind a little bit ;-). Matplotlib won't plot all NaNs images either:

```python import numpy as np import matplotlib.pyplot as plt

a = np.full((2, 2), np.NaN) plt.imshow(a)
plt.show() # raises an error ``` So we have to decide if it's worth to add a bunch of logic in xarray to handle these cases or if it's the user's responsibility to check the data before plotting.

Yes, I agree. We inherited the logic from seaborn, which also has this problem:

```python In [2]: import seaborn as sns In [3]: import numpy as np In [4]: a = np.zeros((2, 2)) * np.NaN In [6]: sns.heatmap(a)

ValueError Traceback (most recent call last) <ipython-input-6-9ef15135ade2> in <module>() ----> 1 sns.heatmap(a)

~/.pyvirtualenvs/py3/lib/python3.5/site-packages/seaborn/matrix.py in heatmap(data, vmin, vmax, cmap, center, robust, annot, fmt, annot_kws, linewidths, linecolor, cbar, cbar_kws, cbar_ax, square, xticklabels, yticklabels, mask, ax, **kwargs) 515 plotter = _HeatMapper(data, vmin, vmax, cmap, center, robust, annot, fmt, 516 annot_kws, cbar, cbar_kws, xticklabels, --> 517 yticklabels, mask) 518 519 # Add the pcolormesh kwargs here

~/.pyvirtualenvs/py3/lib/python3.5/site-packages/seaborn/matrix.py in init(self, data, vmin, vmax, cmap, center, robust, annot, fmt, annot_kws, cbar, cbar_kws, xticklabels, yticklabels, mask) 166 # Determine good default values for the colormapping 167 self._determine_cmap_params(plot_data, vmin, vmax, --> 168 cmap, center, robust) 169 170 # Sort out the annotations

~/.pyvirtualenvs/py3/lib/python3.5/site-packages/seaborn/matrix.py in _determine_cmap_params(self, plot_data, vmin, vmax, cmap, center, robust) 205 calc_data = plot_data.data[~np.isnan(plot_data.data)] 206 if vmin is None: --> 207 vmin = np.percentile(calc_data, 2) if robust else calc_data.min() 208 if vmax is None: 209 vmax = np.percentile(calc_data, 98) if robust else calc_data.max()

~/.pyvirtualenvs/py3/lib/python3.5/site-packages/numpy/core/_methods.py in _amin(a, axis, out, keepdims) 27 28 def _amin(a, axis=None, out=None, keepdims=False): ---> 29 return umr_minimum(a, axis, None, out, keepdims) 30 31 def _sum(a, axis=None, dtype=None, out=None, keepdims=False):

ValueError: zero-size array to reduction operation minimum which has no identity

```

Will try to submit a fix later today