geo=True and rasterize=True makes the plot disappear

[ahuang11]

import xarray as xr
import panel as pn
import holoviews as hv
from holoviews.operation.datashader import rasterize
import geoviews as gv
import cartopy.crs as ccrs

hv.renderer("bokeh").webgl = False
import hvplot.xarray
ds = xr.tutorial.open_dataset("air_temperature").isel(time=0)

rasterize((gv.Image(ds, ["lon", "lat"], ["air"]))) * gv.feature.coastline()

Data disappears on zoom.

[hoxbro]

It is again coming from changes made in holoviz/hvplot#1053

Edit: I was wrong

[ahuang11]

Nice! git bisect again?

[hoxbro]

I was too quick; does not seem to be an hvplot as I can recreate it with:

import xarray as xr
import geoviews as gv
import numpy as np
from holoviews.operation.datashader import rasterize

gv.extension("bokeh")

data = xr.DataArray(
    np.random.random((100, 50)),
    coords=dict(lon=np.arange(200, 300), lat=np.arange(25, 75)),
)

rasterize(gv.Image(data), width=10, height=10)

[hoxbro]

Tried going back in time, and it does not seem to be a recent regression, as no one of the following works:

❯ conda list "python|holoviews|geoviews|bokeh"
# packages in environment at /home/shh/miniconda3/envs/geo_bug:
#
# Name                    Version                   Build  Channel
bokeh                     2.4.3              pyhd8ed1ab_3    conda-forge
brotli-python             1.1.0            py38h17151c0_0    conda-forge
geoviews                  1.9.3              pyhd8ed1ab_0    conda-forge
geoviews-core             1.9.3              pyha770c72_0    conda-forge
holoviews                 1.14.5             pyhd8ed1ab_0    conda-forge
python                    3.8.17          he550d4f_0_cpython    conda-forge
python-dateutil           2.8.2              pyhd8ed1ab_0    conda-forge
python-tzdata             2023.3             pyhd8ed1ab_0    conda-forge
python_abi                3.8                      3_cp38    conda-forge

# packages in environment at /home/shh/miniconda3/envs/geo_bug:
#
# Name                    Version                   Build  Channel
bokeh                     2.2.3            py38h578d9bd_0    conda-forge
brotli-python             1.1.0            py38h17151c0_0    conda-forge
geoviews                  1.8.2                      py_0    conda-forge
geoviews-core             1.8.2                      py_0    conda-forge
holoviews                 1.14.5             pyhd8ed1ab_0    conda-forge
python                    3.8.17          he550d4f_0_cpython    conda-forge
python-dateutil           2.8.2              pyhd8ed1ab_0    conda-forge
python-tzdata             2023.3             pyhd8ed1ab_0    conda-forge
python_abi                3.8                      3_cp38    conda-forge

Will move it to GeoViews for now.

[ahuang11]

What about including datashader?

[hoxbro]

I used Datashader 1.15.2.

I did not expect Datashader to be a problem as it doesn't have any knowledge of geo-related information, and just get coordinates and data passed through it.

Did a test with 0.14.4, which reproduced the same results.

[ahuang11]

I'm quite surprised that no one has commented about this yet besides me haha. I should probably dive deeper into this soon...

[itsgifnotjiff]

I believe i had posted something about it a year or two ago but no one was interested so I deleted the post :) glad to see it ressurface.

[ahuang11]

Workaround seems to work

import xarray as xr
import panel as pn
import holoviews as hv
from holoviews.operation.datashader import rasterize
import geoviews as gv
import cartopy.crs as ccrs

hv.renderer("bokeh").webgl = False
import hvplot.xarray
ds = xr.tutorial.open_dataset("air_temperature").isel(time=0)
# pn.serve(ds.hvplot("lon", "lat", rasterize=True, crs=ccrs.PlateCarree(), projection=ccrs.PlateCarree()))

pn.serve(rasterize(gv.project(gv.Image(ds, ["lon", "lat"], ["air"]))))

[ahuang11]

There are two separate issues here.

1. Bounciness with aspect(?) seen in hvplot:

ds.hvplot("lon", "lat", geo=True, rasterize=True, coastline=True, project=True)

Screen.Recording.2024-04-11.at.3.42.38.PM.mov

No problem in geoviews

import xarray as xr
import panel as pn
import holoviews as hv
from holoviews.operation.datashader import rasterize
import geoviews as gv
import cartopy.crs as ccrs

hv.renderer("bokeh").webgl = False
import hvplot.xarray
ds = xr.tutorial.open_dataset("air_temperature").isel(time=0)

rasterize(gv.project(gv.Image(ds, ["lon", "lat"], ["air"]))) * gv.feature.coastline()

Screen.Recording.2024-04-11.at.3.43.28.PM.mov

2. If project=True isn't set, the data disappears on zoom

import xarray as xr
import panel as pn
import holoviews as hv
from holoviews.operation.datashader import rasterize
import geoviews as gv
import cartopy.crs as ccrs

hv.renderer("bokeh").webgl = False
import hvplot.xarray
ds = xr.tutorial.open_dataset("air_temperature").isel(time=0)

rasterize((gv.Image(ds, ["lon", "lat"], ["air"]))) * gv.feature.coastline()

[ahuang11]

x0 == x1 here for some reason after zooming

    def _process(self, img, key=None):
        from cartopy.img_transform import warp_array

        if self.p.fast:
            return self._fast_process(img, key)

        proj = self.p.projection
        x0, x1 = img.range(0, dimension_range=False)
        y0, y1 = img.range(1, dimension_range=False)
        yn, xn = img.interface.shape(img, gridded=True)[:2]
        (px0, py0, px1, py1) = project_extents((x0, y0, x1, y1), img.crs, proj)

[ahuang11]

Aha, I think it's an issue with 0 to 360 vs -180 to 180

import xarray as xr
import panel as pn
import holoviews as hv
from holoviews.operation.datashader import rasterize
import geoviews as gv
import cartopy.crs as ccrs

hv.renderer("bokeh").webgl = False
import hvplot.xarray
ds = xr.tutorial.open_dataset("air_temperature").isel(time=0)
# 0-360 to -180-180
ds["lon"] = (ds["lon"] + 180) % 360 - 180

rasterize((gv.Image(ds, ["lon", "lat"], ["air"]))) * gv.feature.coastline()

[ahuang11]

Something happening here:

[ahuang11]

https://github.com/holoviz/holoviews/blame/main/holoviews/operation/resample.py#L142

I think it comes down to: x0, x1 is ranged from -180 to 180, while ex0 is ex1 is ranged from 0 to 360, so when the following computed, it gives x_range=(198.75, 198.75)

import numpy as np
 
 
x0, x1, ex0, ex1 = -154.39239111669974, -33.72522063810877, 198.75, 331.25


x_range = (np.nanmin([np.nanmax([x0, ex0]), ex1]),
            np.nanmax([np.nanmin([x1, ex1]), ex0]))
x_range

[ahuang11]

This original issue was an hvplot issue. Renaming this issue for the geoviews issue. holoviz/hvplot#1105