New geometries module - Work In Progress

[AtelierLibre]

Resolves #478

Work in progress on a new geometries module that combines the capabilities of the existing pois and footprints modules and introduces conversion of ways to LineStrings or Polygons as appropriate.

Example:

# New geometries module
MK = ox.gdf_from_point((52.0430,-0.7590), dist=1000, tags={'highway':True, 'landuse':True, 'building':True})

# Presentation code
MK.to_crs(epsg=27700, inplace=True)

fig,ax = plt.subplots(figsize=(16,16))
MK[MK['landuse'].notna()].plot(ax=ax, column='landuse', cmap='tab20b', legend=True)
MK[MK['highway'].notna()].plot(ax=ax, color='white')
MK[MK['building'].notna()].plot(ax=ax, color='dimgrey', zorder=3)
ax.set_facecolor('darkgrey')
ax.axes.xaxis.set_visible(False)
ax.axes.yaxis.set_visible(False)

[AtelierLibre]

@gboeing I think this is getting to the stage where it could do with some thorough testing. I've been doing some myself (and will keep doing more) but it's a bit random and limited in scope so help with that would be much appreciated.

I have generally tried to stay away from the graph module but to get gdf_from_xml() to work I had to make some changes to the _OSMContentHandler. I don't think they should affect graph_from_xml() but might be worth keeping an eye on that. On a related note, I think moving _overpass_json_from_file() and _OSMContentHandler from graph to downloader would help to make clear that they are shared with geometries.

Let me know what you think/what's the best way forward.

[gboeing]

Fantastic. Thank you. I'm grateful for all the hard work you've put in. Perhaps we should merge this PR into a feature branch here, when you're comfortable with it as a "first draft"? And then start some coordinated testing?

On a related note, I think moving _overpass_json_from_file() and _OSMContentHandler from graph to downloader would help to make clear that they are shared with geometries.

I think that's fine.

I think this is getting to the stage where it could do with some thorough testing.

Certainly. Let's do something like this. I'll work up a big processing batch to start downloading geometries all over the world, passing in various tag combos. My goal will be to look for exceptions: do we find anything that breaks the code. I'll aim for several thousand requests sampled across urban areas worldwide.

In the meantime, it may be most efficient if your testing focuses on validation... inspecting if individual queries conform to expectations, given the data on OSM. @xgerrmann would you have any appetite for validation-style testing too?

[AtelierLibre]

No problem, it's nice to have the opportunity to contribute, and it will be good to check that it all works as expected.

Perhaps we should merge this PR into a feature branch here, when you're comfortable with it as a "first draft"? And then start some coordinated testing?

That sounds good. I think that makes a lot of sense. I'll keep going with some smaller scale tests for a couple more days and then assuming the outputs are as expected it will be great to scale it up and check the robustness.

[AtelierLibre]

I have been working on a few tests which are in notebooks here:

https://github.com/AtelierLibre/osmnx_tests/tree/master/notebooks

The notebooks are a bit rough but the results are standing up so far. I'll do a few more and then we can broaden it out.

[AtelierLibre]

@gboeing I've run through quite a few tests and documented them in the notebooks linked to above. It's seeming pretty robust.

The comparisons with overpass-turbo.eu were particularly helpful as they give something external to benchmark against in terms of numbers. I haven't got to the bottom of the discrepancy with the boundary polygon from the nominatim api compared to the boundary polygon through the overpass api on the final example in that notebook.

The XMLs have thrown up a few errors mainly because it doesn't seem guaranteed that they are complete (unlike the responses from the overpass api as far as I have found). I've added in a few try: except: to catch those that have come up so far.

Let me know if there is anything else you think is worth checking, otherwise I think it probably makes sense to start some wider scale testing now.

[gboeing]

Ok. I'm going to merge this into a geometries feature branch. Testing and additional development can continue there.

[gboeing]

See #549