Norm keywords aren't passing properly to colormap

[bradyrx]

I'm finding that the keywords for e.g. LogNorm are not passing through the wrapper to matplotlib appropriately. Using the same data as when we discussed the LogNorm colorbar in #25: http://s000.tinyupload.com/index.php?file_id=02946064711977690771.

Below, see that proplot can't modify the LogNorm boundaries.

With proplot:

f, ax = plot.subplots(proj='pcarree', tight=False, axwidth=6,)

p = ax.contourf(chl.lon, chl.lat, chl.values, norm='log', norm_kw={'vmin': 0.01, 'vmax': 10}, levels=30,
                cmap='Fire', )

ax.format(latlim=(20, 50), lonlim=(-140, -105), land=True)
ax.colorbar(p, loc='r')

With matplotlib:

f, ax = plt.subplots()
p = plt.pcolormesh(chl.lon, chl.lat, chl.values, norm=colors.LogNorm(vmin=0.01, vmax=1), cmap='Fire')
plt.colorbar(p)

[lukelbd]

If you pass vmin and vmax to contourf directly, they will be used:

ax.contourf(chl.lon, chl.lat, chl.values, norm='log', vmin=0.01, vmax=10, levels=30, cmap='Fire', ) 

But it's definitely not right to just ignore norm_kw. Just pushed a patch (5b09d89), now cmap_wrapper checks to see if vmin and vmax were passed in the norm_kw dictionary.

[bradyrx]

@lukelbd, FYI, this solution doesn't work. I forgot I tried that before.

plot.rc['geogrid.alpha'] = 0

plot.rc.fontname = 'Helvetica Light'
f, ax = plot.subplots(proj='cyl', axwidth='8cm',)
p = ax.contourf(chl.lon, chl.lat, chl, norm='log', vmin=0.01, vmax=10, levels=30)
ax.colorbar(p, loc='r')
ax.format(land=True, latlim=(20, 50), lonlim=((-140, -105)))

I think you might get this behavior if you use the LogLocator directly. (https://matplotlib.org/3.1.1/gallery/images_contours_and_fields/contourf_log.html)

[lukelbd]

Okay this patch seems to do the trick: a3ebbb1. Try:

f, ax = plot.subplots()
ax.pcolormesh(10**(np.random.rand(10,10)*5), vmin=0.1, vmax=10, norm='log', colorbar='r')

Previously if you specified both vmin and vmax I just set the levels to np.linspace(vmin, vmax, N), because most locators treat vmin and vmax as suggested locations, not absolute locations. For example, if you used vmax=20, the actual level maximum would be 10. But now this is documented, and I think it makes more sense that if you really want exact locations, you should provide your own levels array.