ENH: mapclassify plot legend as discrete colorbar

[TLouf]

fixes #2154

This entails a change in the fmt keyword to format tick labels, as colorbars' tick labels have been formatted with old-style formats (like "%.3f"), until the very recent matplotlib/matplotlib/pull/21542. By the way I wonder if it is desirable to keep "%.2f" as default formatting, or to rather not provide one.

I left a TODO about adding a legend entry for geometries with missing data, which I think would be easier to implement in #2012 once this one is in.

Example output:

import geopandas
from geopandas.datasets import get_path
world = geopandas.read_file(get_path('naturalearth_lowres'))
world.plot(column="pop_est", scheme="BoxPlot", legend=True)

[martinfleis]

I personally like this change but I think it needs a wider discussion as it is essentially a breaking change. I think that in future, we should default to colorbar but I am not sure that we should do that abruptly, so maybe some transitional period would be helpful.

And I think that we should make sure that colorbar is not larger than the plot itself.

[Cloudiver]

I totally agree with @TLouf's suggestion. In many research paper, when different areas are distinguished by various colors, they are usually discrete colorbar, in order to highlight the difference because of continuous colorbar are not easy to read specific values. In other cases it may be to show the overall trend between regions, in this case a continuous colorbar is usually used.

Overall, I wonder if it's possible to have this feature as an optional parameter to plot instead of suddenly being the default colorbar.

[TLouf]

I agree with the optional parameter, I restored the historical behaviour and set it as default. A colorbar can be generated by passing as_cbar=True in legend_kwds.

And I think that we should make sure that colorbar is not larger than the plot itself.

Personally I don't think so, this would imply using AxesDivider for instance, which also has its keyword arguments. So instead of setting an arbitrary default I'd go with matplotlib's, although it results in this height issue when the aspect ratio is not equal to 1. Then it'd be up to the user, as in matplotlib, to create a cax of their own using AxesDivider or something else.

[Yimsun97]

I am recently working on the discrete colorbar in my research program but I am no expert in the underlying implementation and combination of geopandas with matplotlib. I also find that the gridded maps with the same or different colorbar(s) are widely used in many papers. Will the geopandas offer options for flexible colorbar creation (e.g. the single colorbar for all subplots and one colorbar for each subplot) in the future? I came up with a workaround script by using the "UserDefined" scheme with self-defined colorbar, but it may not be applicable for general cases. P.S. I am a fan of geopandas and would like to hear about update on this functionality.