BUG: Incorrect colours when plotting multiple geometry types with list of colours

Expected result

I have a geopandas.GeoDataFrame that contains multiple geometry types (e.g. Polygon, LineString and Point). I want to plot this data by supplying a list of colours (one per feature) to the color= parameter:

from shapely.geometry import LineString, Point, Polygon
import geopandas as gpd

gdf = gpd.GeoDataFrame(geometry=[Polygon([(1, 1), (1, 2), (2, 2), (2, 1)]),
                                LineString([(2, 0), (4, 2)]),
                                Point(3, 2),
                                Point(1, 0)])

gdf.plot(color=['red', 'green', 'blue', 'blue'])

For this example, I would expect that:

1. My Polygon feature will be plotted in “red”
2. My LineString feature will be plotted in “green”
3. Both of my Point features will be plotted in “blue”

Actual result

However, I receive this result:

It appears to me like the list of colours is being applied separately to each geometry type, with the first item of each type being plotted in “red”, and the second item of each type being plotted in “green” (e.g. see the Point features).

This appears related to the PR and linked issues here (#1119), but those look like they refer specifically to multi-geometries, and not multiple geometry types.

Environment geopandas '0.7.0' Python 3.6.9 (default, Nov 7 2019, 10:44:02) IPython 7.12.0

[martinfleis]

Thank you for the report!

Your assumption is correct. Each geometry type is now plotted as an independent Collection, to which we pass color as it is (assuming it is a single value). For list-like color values of the correct length, we should do the same we do with value, i.e. decomposing the list based on the geom type.

[martinfleis]

could be considered an misunderstanding of color parameter. It’s the sequence in which colors are used based on column parameter.

Not necessarily. You can plot geometries without using a column and then the idea is that you are able to pass a list of colours that is mapped to individual geometries. It is just broken when you have mixed geom types.

if length of color parameter is same as dataframe, create an internal column and a LinearSegmentedColormap() as per examples above

That may work, yes. You don’t even need a column as column can take an array.

But the solution I outlined above would also work:

For list-like color values of the correct length, we should do the same we do with value, i.e. decomposing the list based on the geom type.