Geoviews not showing maps, data upside down

RicceB · August 17, 2021, 9:20am

Hi,

I recently installed geoviews on a server, to be used together with JupyterLab. Installation following the instructions from geoviews.org work without problem. Then I tried to reproduce the airports graph but I did not get the right output.

The first cell:

import pandas as pd
import geoviews as gv

from geoviews import opts
from bokeh.sampledata.airport_routes import airports, routes

gv.extension('bokeh')

works without any problem. So does the second cell:

# Count the number of connections from each airport
counts = routes.groupby('SourceID')[['Stops']].count().reset_index().rename(columns={'Stops': 'Connections'})
airports_df = pd.merge(airports, counts, left_on='AirportID', right_on='SourceID', how='left')

# Select only US mainland airports & convert from Mercator to Latitudes/Longitudes
airport_points = gv.Points(airports_df, ['Longitude', 'Latitude']).select(Longitude=(-170, -50), Latitude=(0, 50))

# Declare nodes, graph and tiles
nodes = gv.Nodes(airport_points, ['Longitude', 'Latitude', 'AirportID'],
             ['Name', 'City', 'Connections'])
graph = gv.Graph((routes, nodes), ['SourceID', 'DestinationID'], ['Source', 'Destination'])
tiles = gv.tile_sources.Wikipedia

# Select 50 busiest airports
busiest = list(routes.groupby('SourceID').count().sort_values('Stops').iloc[-50:].index.values)
busiest_airports = graph.select(AirportID=busiest, selection_mode='nodes')

The third cell runs without any problem but the output is not as expected:

(tiles * busiest_airports).opts(
    opts.Graph(edge_selection_line_color='black', edge_hover_line_color='red',
               edge_line_width=1, edge_line_alpha=0.01, edge_nonselection_line_alpha=0.01,
               width=800, height=600))

A graphical object is displayed but it looks empty. The zoom is not automatically set, so I need to zoom out manually (quite a bit, in fact). When I finally find the data, it is plotted with mirrored axis (latitude increasing from top to bottom, longitude decreasing from left to right), as you can see in the image below. The map does not appear either.

I originally thought that the problem could be due to some connection being blocked somewhere, but having the axis mirrored made me think this may be a different issue. When I try the code on a local installation of geoviews (that normally works fine) without internet connection the map is not displayed but the axis are not mirrored.

Regarding package versions, when I installed geoviews, the following packages were installed:

ca-certificates  {2021.5.30 (conda-forge/linux-64) -> 2021.7.5 (defaults/linux-64)}
certifi  {2021.5.30 (conda-forge/linux-64) -> 2021.5.30 (defaults/linux-64)}
conda  {4.10.1 (conda-forge/linux-64) -> 4.10.3 (defaults/linux-64)}
krb5  {1.18.2 (defaults/linux-64) -> 1.19.2 (defaults/linux-64)}
libcurl  {7.71.1 (defaults/linux-64) -> 7.78.0 (defaults/linux-64)}
libedit  {3.1.20191231 (defaults/linux-64) -> 3.1.20210216 (defaults/linux-64)}
sqlite  {3.33.0 (defaults/linux-64) -> 3.36.0 (defaults/linux-64)}
+bokeh-2.3.3 (defaults/linux-64)
+cairo-1.16.0 (defaults/linux-64)
+cartopy-0.18.0 (defaults/linux-64)
+cfitsio-3.470 (defaults/linux-64)
+cftime-1.5.0 (defaults/linux-64)
+click-7.1.2 (defaults/noarch)
+click-plugins-1.1.1 (defaults/noarch)
+cligj-0.7.2 (defaults/linux-64)
+colorcet-2.0.6 (pyviz/noarch)
+curl-7.78.0 (defaults/linux-64)
+cytoolz-0.11.0 (defaults/linux-64)
+dask-2021.7.2 (defaults/noarch)
+dask-core-2021.7.2 (defaults/noarch)
+datashader-0.13.0 (pyviz/noarch)
+datashape-0.5.4 (defaults/linux-64)
+distributed-2021.7.2 (defaults/linux-64)
+fiona-1.8.13.post1 (defaults/linux-64)
+freexl-1.0.6 (defaults/linux-64)
+fsspec-2021.7.0 (defaults/noarch)
+gdal-3.0.2 (defaults/linux-64)
+geopandas-0.9.0 (defaults/noarch)
+geopandas-base-0.9.0 (defaults/noarch)
+geos-3.8.0 (defaults/linux-64)
+geotiff-1.6.0 (defaults/linux-64)
+geoviews-1.9.1 (defaults/noarch)
+geoviews-core-1.9.1 (pyviz/noarch)
+giflib-5.2.1 (defaults/linux-64)
+hdf4-4.2.13 (defaults/linux-64)
+hdf5-1.10.6 (defaults/linux-64)
+heapdict-1.0.1 (defaults/noarch)
+holoviews-1.14.5 (pyviz/noarch)
+json-c-0.13.1 (defaults/linux-64)
+kealib-1.4.14 (defaults/linux-64)
+libdap4-3.19.1 (defaults/linux-64)
+libev-4.33 (defaults/linux-64)
+libgdal-3.0.2 (defaults/linux-64)
+libkml-1.3.0 (defaults/linux-64)
+libnetcdf-4.6.1 (defaults/linux-64)
+libnghttp2-1.41.0 (defaults/linux-64)
+libpq-12.2 (defaults/linux-64)
+libspatialindex-1.9.3 (defaults/linux-64)
+libspatialite-4.3.0a (defaults/linux-64)
+locket-0.2.1 (defaults/linux-64)
+mapclassify-2.4.3 (defaults/noarch)
+markdown-3.3.4 (defaults/linux-64)
+msgpack-python-1.0.2 (defaults/linux-64)
+multipledispatch-0.6.0 (defaults/linux-64)
+munch-2.5.0 (defaults/noarch)
+netcdf4-1.5.7 (defaults/linux-64)
+networkx-2.6.2 (defaults/noarch)
+openjpeg-2.3.0 (defaults/linux-64)
+panel-0.12.1 (pyviz/noarch)
+param-1.11.1 (pyviz/noarch)
+partd-1.2.0 (defaults/noarch)
+pixman-0.40.0 (defaults/linux-64)
+poppler-0.81.0 (defaults/linux-64)
+poppler-data-0.4.10 (defaults/linux-64)
+postgresql-12.2 (defaults/linux-64)
+proj-6.2.1 (defaults/linux-64)
+psutil-5.8.0 (defaults/linux-64)
+pyct-0.4.8 (pyviz/noarch)
+pyct-core-0.4.8 (pyviz/noarch)
+pyproj-2.6.1.post1 (defaults/linux-64)
+pyshp-2.1.3 (defaults/noarch)
+pyviz_comms-2.1.0 (pyviz/noarch)
+pyyaml-5.4.1 (defaults/linux-64)
+rtree-0.9.7 (defaults/linux-64)
+shapely-1.7.1 (defaults/linux-64)
+sortedcontainers-2.4.0 (defaults/noarch)
+tblib-1.7.0 (defaults/noarch)
+tiledb-2.2.9 (defaults/linux-64)
+toolz-0.11.1 (defaults/noarch)
+typing-extensions-3.10.0.0 (defaults/noarch)
+typing_extensions-3.10.0.0 (defaults/noarch)
+xarray-0.19.0 (defaults/noarch)
+xerces-c-3.2.3 (defaults/linux-64)
+zict-2.0.0 (defaults/noarch)

Does anyone know what can be wrong or where to start looking? I have not been able to find much information online.

Thanks!

/RicceB

ahuang11 · August 17, 2021, 2:12pm

Maybe specify crs=?

RicceB · August 17, 2021, 8:19pm

Thanks for the tip. I tried changing one line in the second cell to include crs=, like this:

airport_points = gv.Points(airports_df, ['Longitude', 'Latitude'], crs=crs.PlateCarree()).select(Longitude=(-170, -50), Latitude=(0, 50))

but this did not change anything. However, your reply made me think that perhaps there was an issue with projection= rather than with crs=. So I tried changing the third cell to the following:

(tiles * busiest_airports).opts(
opts.Graph(edge_selection_line_color='black', edge_hover_line_color='red',
           edge_line_width=1, edge_line_alpha=0.01, edge_nonselection_line_alpha=0.01,
           width=800, height=600, projection=crs.Mercator()))

This made things work. The strange thing is that if after having executed the command above I execute the original command, that is without projection=crs.Mercator(), it works as expected. I cannot understand why.

[Although the issue is now somewhat solved, let me also say that I tried using matplotlib as backend and things worked as well most times. For example, the following command produced the expected output

gv.output(busiest_airports, backend='matplotlib', fig='svg', size=300)

Of course without a map].

RicceB · August 18, 2021, 9:11am

One further update.

It seems that the above comment did not totally solve the issue. When zooming in quite a bit, I noticed that the points were slightly shifted with respect to their true positions.

I quickly concluded that I should be using projection=GOOGLE_MERCATOR instead of projection=crs.Mercator() but things did not work when I tried the following:

buff = (tiles * busiest_airports).opts(
    opts.Graph(edge_selection_line_color='black', edge_hover_line_color='red',
               edge_line_width=1, edge_line_alpha=0.01, edge_nonselection_line_alpha=0.01,
               width=800, height=600, projection=crs.GOOGLE_MERCATOR))
buff

Again all I saw was the data upside down with no map. I tried changing the version of cartopy as follows:

The following packages will be DOWNGRADED:

  cartopy                             0.18.0-py38h0d9ca2b_1 --> 0.17.0-py38hee755e7_1

but this did not work either. I did not dig in further. Given that crs.GOOGLE_MERCATOR is just a specific instance of a Mercator projection, I copied the definition from crs.py in cartopy:

from cartopy._crs import (CRS, Geodetic, Globe, PROJ4_VERSION)

WGS84_SEMIMAJOR_AXIS = 6378137.0
WGS84_SEMIMINOR_AXIS = 6356752.3142

GM = crs.Mercator(min_latitude=-85.0511287798066,
                           max_latitude=85.0511287798066,
                           globe=Globe(ellipse=None,
                                       semimajor_axis=WGS84_SEMIMAJOR_AXIS,
                                       semiminor_axis=WGS84_SEMIMAJOR_AXIS,
                                       nadgrids='@null'))

That made things work, so that the following command produces the desired map with the data in the right place:

(tiles * busiest_airports).opts(
    opts.Graph(edge_selection_line_color='black', edge_hover_line_color='red',
               edge_line_width=1, edge_line_alpha=0.01, edge_nonselection_line_alpha=0.01,
               width=800, height=600, projection=GM))

As explained earlier, this seems to change the behavior of earlier commands. In fact, if we simply display what we stored in buff, then it looks as it should. It is not even necessary to execute again the command that assigns the value to buff. It seems to change the behavior for displaying the data.

ahuang11 · August 18, 2021, 4:29pm

Maybe you can submit a geoviews issue.

RicceB · August 19, 2021, 8:59am

Done: Problems with crs.GOOGLE_MERCATOR · Issue #519 · holoviz/geoviews (github.com)

Thanks for your help.

/RicceB