Access GeoPandas points' lat/long in encode() ?

Hi,

Is there an elegant way to adress the lat and long component of points in the geometry column of a Geopandas Dataframe in Altair’s encode method to link it to latitude and longitude? I could not find anything on this in the documentation.

I understand that it transparently does it for mark_geoshape(), but for the case of graduated symbol maps, e.g., via mark_circle(), it would be nice to have a bit more control and not have to generate separate columns in the DataFrame just to point at it.

Here is a relatively minimal example:

import altair as alt
import pandas as pd
import geopandas as gpd

districts = gpd.read_file("https://opendata.potsdam.de/explore/dataset/statistische-bezirke-in-potsdam/download/?format=geojson")
schools = gpd.read_file("https://opendata.potsdam.de/explore/dataset/schulen/download/?format=geojson")

# i would like to skip this step:
schools['lon'] = schools['geometry'].x
schools['lat'] = schools['geometry'].y

basemap = alt.Chart(districts).mark_geoshape(fill="lightgray", stroke="darkgray"
                    ).encode(tooltip="bezeichnun")

pois = alt.Chart(schools).mark_circle().encode(
    latitude='lat:Q',
    longitude='lon:Q',
    tooltip="schulname",
    size="planungsra:Q"
)

basemap + pois

Thanks a lot, Marian

[jakevdp]

Actually, I take that back. There is a shortcut here, using vega-lite’s syntax for nested data encodings:

import altair as alt
import pandas as pd
import geopandas as gpd

districts = gpd.read_file("https://opendata.potsdam.de/explore/dataset/statistische-bezirke-in-potsdam/download/?format=geojson")
schools = gpd.read_file("https://opendata.potsdam.de/explore/dataset/schulen/download/?format=geojson")

basemap = alt.Chart(districts).mark_geoshape(fill="lightgray", stroke="darkgray"
                    ).encode(tooltip="bezeichnun")

pois = alt.Chart(schools).mark_circle().encode(
    latitude='geometry.coordinates[1]:Q',
    longitude='geometry.coordinates[0]:Q',
    tooltip="schulname",
    size="planungsra:Q"
)

basemap + pois

[mattijn]

Yeah also just realized that it should be possible like this. It’s a bit better then assigning new lat, lon columns👍.

Side note: You could not have known this from the Python side, since there is no schools.geometry.coordinates in the GeoDataFrame. The GeoDataFrame is serialised as (slightly custom) GeoJSON for Vega-Lite insertion and referred to in the Altair spec.