Investigate the effects of low topoquantization epsilon values

[mattijn]

Investigate the effects of low topoquantization epsilon values

[mattijn]

I'm happy this issue is resolved. Its fun to play with this.

import topojson as tp
import geopandas as gpd

world = gpd.read_file(gpd.datasets.get_path("naturalearth_lowres"))
tp.Topology(data=world).toposimplify(4).topoquantize(20).to_alt(color='properties.name:N')

[mattijn]

I've to admit it's not fast.

import topojson as tp
import geopandas as gpd
import matplotlib.pyplot as plt

brexit = gpd.read_file('/Users/mattijnvanhoek/Downloads/brexit.gpkg')
topo = tp.Topology(brexit)

q100 = topo.topoquantize(100).to_gdf()
q250 = topo.topoquantize(250).to_gdf()
q500 = topo.topoquantize(500).to_gdf()

f, ax = plt.subplots(1, 3, figsize=(10, 10))
q100.plot(ax=ax[0])
q250.plot(ax=ax[1])
q500.plot(ax=ax[2])

Would indeed be good to investigate routes for quantization through pygeos as well, like in this comment geopandas/geopandas#1727 (comment)

[ljwolf]

We're seeing these issues as well.

brexit.gpkg.zip

Interested in digging into this...

[mattijn]

Thanks for your interest in this issue. At this place in the docs https://mattijn.github.io/topojson/example/settings-tuning.html#prequantize I've worked out the quantization process in 1D.

I see you have linked this issue to another at the geopandas repo. Im still figuring out what is going wrong here. Maybe your usecase is a smaller reproducible example.

[mattijn]

Found a use-case small enough for debugging:

import topojson as tp
import geopandas as gpd

gdf = gpd.read_file(gpd.datasets.get_path("naturalearth_lowres"))
gdf = gdf[gdf.name.isin(['France', 'Belgium', 'Netherlands'])]
tp.Topology(data=gdf).toposimplify(4).topoquantize(500).to_alt(color='properties.name:N')

[mattijn]

Thanks again for commenting in order to bring this issue on my radar again!

This issue is fixed by #116, linestrings which can become points due filtering of repeating coordinates that were not written back to the linestring. After delta-encoding the start-coordinates of quantised linestrings got mixed up with this case.

During debugging I found out that the overal bbox is not regenerated after simplification and quantization. Filled #117 for this.

[ljwolf]

If you're game for a numba implementation, I'm sure that'd help accelerate things? I prefer working on the topojson-style data structure, and numba could work well with ragged arrays stored as lists of tuples.

[mattijn]

This is discussed previously with @Casyfill and might be a good method indeed to improve speed further. It might be very simple to get good results, but I've never tried it. Let me open a new issue so it can be tracked.