Spatial bias plot error with mpl 3.6

[zmoon]

Got error message: unhashable type: 'ListedColormap' in spatial bias plot

Abbreviated traceback:

monet.plots.sp_scatter_bias(..., cmap=new_color_map(), ...) passes a custom ListedColormap.
In monet, this is passed through to pandas df.plot.scatter

Based on the message below, it seems pandas is expecting a cmap name if mpl >= 3.6?

│ /scratch/zmoon/mambaforge/envs/mm/lib/python3.9/site-packages/pandas/plottin │
│ g/_matplotlib/core.py:1225 in _make_plot                                     │
│                                                                              │
│   1222 │   │                                                                 │
│   1223 │   │   if self.colormap is not None:                                 │
│   1224 │   │   │   if mpl_ge_3_6_0():                                        │
│ > 1225 │   │   │   │   cmap = mpl.colormaps[self.colormap]                   │
│   1226 │   │   │   else:                                                     │
│   1227 │   │   │   │   cmap = self.plt.cm.get_cmap(self.colormap)            │
│   1228 │   │   else:

[zmoon]

For anyone who needs a quick fix, conda install 'matplotlib-base <3.6'. But I will work on resolving this for mpl 3.6.

[jianheACM]

@zmoon @rschwant I had the same issue after updating the pkg.

Below is my quick fix:
In function new_color_map

...
    newcmap = ListedColormap(newcolors, name='OrangeBlue')
    mpl.colormaps.register(newcmap)

    return newcmap

In spatial bias plot:

...
    newcmap = 'OrangeBlue'
    try:
        mpl.colormaps[newcmap]
    except KeyError:
        new_color_map()
...

This will work for matplotlib >=3.6

[zmoon]

@jianheACM the bug in pandas was fixed in v1.5.2. What is your pandas version?

[zmoon]

Registering and using the string name seems like a good workaround though, maybe we should add this to MM anyway.

[jianheACM]

It's v1.5.1. But it's good to know it is fixed in v1.5.2.
I feel like we will have some similar issues in the future if we update pkg version.

[rschwant]

@jianheACM and @zmoon if you feel like registering the colormap like Jian did above will lead to less package versioning incompatibilities, I'm fine with updating this like Jian suggests.

[rschwant]

A better method for this has been merged to develop, to avoid future problems. Since this doesn't cause problems currently I am closing it.
#154