ENH: plot_density for KDE plotting of point patterns based on statsmodels

[martinfleis]

Resolves https://github.com/gdsbook/book/issues/305. Essentially replaces seaborn.kdeplot which does not work as expected with our own solution.

Tests will come once I'll figure out how to do them.

import geopandas as gpd
from pointpats.kde import plot_density

gdf = gpd.read_file('https://martinfleischmann.net/sds/chapter_06/data/brno_pedestrian_accidents.gpkg')

gdf.plot(markersize=0.5, figsize=(9,9))
plot_density(gdf, 500)

gdf.plot(markersize=0.5, figsize=(9,9))
plot_density(gdf, 500, levels=25, cmap="magma_r", linewidths=1)

gdf.plot(markersize=0.1, figsize=(9,9), zorder=2, color='k')
plot_density(gdf, 500, levels=25, cmap="magma_r", fill=True)

gdf.plot(markersize=0.5, figsize=(9,9))
plot_density(gdf, 1000, cmap="magma_r", linewidths=1)

I would love to have this in the next release alongside #111 so I can use both during teaching starting in two weeks.

[codecov[bot]]

Codecov Report

Merging #118 (26b45b3) into main (8fd02a6) will increase coverage by 0.7%. The diff coverage is 86.4%.

@@           Coverage Diff           @@
##            main    #118     +/-   ##
=======================================
+ Coverage   63.5%   64.2%   +0.7%     
=======================================
  Files         11      12      +1     
  Lines       1404    1448     +44     
=======================================
+ Hits         892     930     +38     
- Misses       512     518      +6     

Files Changed	Coverage Δ
pointpats/kde.py	86.0% <86.0%> (ø)
pointpats/__init__.py	100.0% <100.0%> (ø)

[knaaptime]

funny. I'm also right in the middle of using this for some bike/ped accident work

[martinfleis]

I was struggling to make it work decently fast on large data so I included also an implementation on top of KDEpy. That is fast, offers a selection of kernels but can behave funny in some edge cases. While statsmodels seems to be robust default.

[martinfleis]

The tests are not perfect but testing matplotlib is just a pain in the ass, so this at least ensures that the stuff works and args are used. It should be ready for review.

[knaaptime]

right now this gives back a matplotlib.contour.QuadContourSet. Is there any way to have it return an axes object, e.g. to play nicely with contextily?