avoid radius estimation when using "delaunay_triangulation"

[domagojhack]

I am successfully embedding EOmaps into my app. However, I have a slight "hiccup". My App will work with a lot of raster data.
However, each time plot is redrawn I see that EOmaps conduct something that is called Radius estimation. (It takes about a second or two per raster layer). As I am working with data that will have more and less the same resolution, extent etc.
It looks like this Radius estimation could be circumvented using some precalculated radius value.

I tried to look in the source but could not trace the radius argument, only the estimator.

Here is a minimum code example:

import geopandas as gpd
from eomaps import Maps
import matplotlib
%matplotlib widget

m = Maps(layer="DEM")

data = m.read_file.GeoTIFF(
    "base_layers/dem/myDEM.tif")
m.set_shape.delaunay_triangulation()
m.set_data(**data)
m.plot_map(cmap="gist_earth")

Output I am getting:

EOmaps: estimating radius...
EOmaps: radius: 12.5000

How can I set radius so I can circumvent radius estimation?

Cheers

[raphaelquast]

Hey, sounds awesome! I'd love to get a look at it!

For delaunay-triangulations the radius-estimation is used differently than for other shapes...
The radius is only used as a "mask_radius" to mask the obtained triangulation
(to avoid drawing very large triangles that connect far-away points).
If the mask-radius is not provided explicitly EOmaps tries to come up with a suitable guess based on the pixel-spacing (hence the radius-estimation).

To avoid the estimation in the current release, simply use m.set_shape.delaunay_triangulation(mask_radius=np.inf) (or use any number appropriate to your data). Not sure though if the radius-estimation is the slow part here... I'd expect the evaluation of the triangulation to take up most of the time...

I'm actually in the process of improving radius-estimation a little bit (see #87)...
In EOmaps >= v4.2.2 m.set_shape.delaunay_triangulation(masked=False) will also avoid the associated radius estimation step.

[domagojhack]

Here is a sneak peek!

Great! Yeah I've got just a light speedup from m.set_shape.delaunay_triangulation(mask_radius=np.inf) probs only from the lack of print statements. Thanks for the info!

[raphaelquast]

Wow, that looks really nice! I didn't look a lot into embedding EOmaps in another gui so far but I'll definitely have to get a closer look on wxpython if I find the time for it 😄 (I especially like the fact that it just wraps system-ui elements... gets you started a lot quicker I guess)

• If you're datasets are large and speed is of concern, shapes such as "raster", "shade_raster" or "shade_points" should be a lot faster than Delaunay triangulations.
• ... and just to put it here as well.. If you intend to distribute the app on a broader scale using OSM-tiles, make sure to comply to their license! (OSM is free but heavy server-bandwidth usage is not... https://operations.osmfoundation.org/policies/tiles/ )

... also If by any chance you'd be willing to share some of your insights on the wxpython-embedding, I'd very much appreciate a section in the documentation explaining the basics on how to "Embed EOmaps in a wxpython gui" (maybe even with a minimal example...).

[domagojhack]

I don't use OSM-tiles per se, I am using offline OSM data I downloaded and extracted from daily-updated OSM files from http://download.geofabrik.de/. It is much faster and does not get affected by server-bandwidth limitations while being enough for my use case. (I plan, in the future, to host my own osm tile server for apps). It is a bit tedious to prepare data in a format that can be read using Geopandas but it works.

I am more than willing, I already started to work on the documents about embedding in wxpython and I am also researching possible embedding in Kivy framework as well.

[raphaelquast]

OK, that's of course the optimal way to do it 🙂 (not tested but it seems they offer shapefiles as well which gpd.read_file() should be able to read right away... (however for large amounts of polygons it tends to get quite slow)

I am more than willing, I already started to work on the documents about embedding in wxpython and I am also researching possible embedding in Kivy framework as well.

Nice, really looking forward to this! Let me know if you need any help!
Kivy sure looks interesting as well!

[domagojhack]

They provide vector data indeed but in form of binary *.osm.pbf data format, not ESRI shapefiles. I managed to extract and transform data for my ROI, into something smaller, manageable, and readable with Geopandas. So the polygon number is not too big.