hvplot.image() zoom causes axes limits to bounce around

[scottyhq]

Thanks for contacting us! Please read and follow these instructions carefully, then delete this introductory text to keep your issue easy to read. Note that the issue tracker is NOT the place for usage questions and technical assistance; post those at Discourse instead. Issues without the required information below may be closed immediately.

ALL software version info

hvplot 0.5.2
bokeh 2.0.1

Description of expected behavior and the observed behavior

Plotting image results in axes and colorbar limits jumping around. A zoom box draw partially outside valid pixels of image causes further jumping around of axes limits.

Complete, minimal, self-contained example code that reproduces the issue

https://gist.github.com/scottyhq/0e245fcd62d8a46b82a4186fcd8945e6#file-hvplot-zoom-ipynb

# Lots of window and colorbar rescaling happening when first plot, and upon zooming
band1.hvplot.image(x='x',y='y',
                   crs='epsg:32645', 
                   tiles='OSM', 
                   rasterize=True,
                   width=700, height=500, 
                   cmap='plasma', 
                   logz=True,
                   alpha=0.7)

Stack traceback and/or browser JavaScript console output

No error or traceback, but the behavior is buggy. See gif:

Screenshots or screencasts of the bug in action

[rsignell-usgs]

We this issue also. For us it was not bouncing but constant unzooming! intake/xrviz#70 (comment)

[philippjfr]

Can't seem to reproduce anymore with latest versions of everything. Will make dev versions of everything available for testing.

[scottyhq]

@philippjfr - This still seems to be an issue with the latest hvplot release (0.6). I updated the binder link in the first comment with new versions, so that you can reproduce the issue. The bug seems to be limited to when crs is set as an argument.

[philippjfr]

I think this won't be fixed properly until I release HoloViews 1.13.3. Hopefully this weekend.

[scottyhq]

Related issue: holoviz/geoviews#418

[philippjfr]

I can still reproduce this but tackling this now before I release 1.13.3. Note that a workaround for now is to set a frame_width and frame_height. The problem is that the initial aspect calculations are off because the regular width/height include the axes and colorbars, which means that once it knows the size of the canvas it makes a number of adjustments and gets into weird loops because the width of tick labels changes which again causes the canvas area to change in size causing another change in the aspect and so on.

[rsignell-usgs]

@philippjfr, I'm still getting this issue with holoviews 1.13.3 in certain cases (although I can fix it by setting frame_width and frame_height as you suggest).

This video demonstrates the problem.

The code to reproduce the video is:

import xarray as xr
import hvplot.xarray

url = ('http://geoport.whoi.edu/thredds/dodsC/vortexfs1/usgs/users'
      '/scook/proj/DE_Bay/run07/subtidal/de_bay_run07_subtidal.ncml')

ds = xr.open_dataset(url)
ds.salt.sel(ocean_time='2017-08-05')[-1,:,:].hvplot.quadmesh(x='lon_rho',y='lat_rho', 
            geo=True, tiles='OSM', rasterize=True, cmap='rainbow')

And then change the value using the selection slider.

[rsignell-usgs]

Ugh. I'm getting it also even if I specify the frame_width and frame_height:

https://www.screencast.com/t/qyLnryCbbv

import xarray as xr
import hvplot.xarray

url = ('http://geoport.whoi.edu/thredds/dodsC/vortexfs1/usgs/users'
      '/scook/proj/DE_Bay/run07/subtidal/de_bay_run07_subtidal.ncml')

ds = xr.open_dataset(url)
ds.salt.hvplot.quadmesh(x='lon_rho',y='lat_rho', frame_width=600, frame_height=400,
            geo=True, tiles='OSM', rasterize=True, cmap='rainbow')

[philippjfr]

Strange, I can definitely reproduce.

[philippjfr]

@rsignell-usgs The dataset doesn't seem to be available anymore: http://geoport.whoi.edu/thredds/dodsC/vortexfs1/usgs/users/scook/proj/DE_Bay/run07/subtidal/de_bay_run07_subtidal.ncml

Is there anywhere I can access it?

[rsignell-usgs]

@philippjfr , the link you provided is the OPeNDAP Data URL -- you need to append .html to browse the dataset metadata so:
http://geoport.whoi.edu/thredds/dodsC/vortexfs1/usgs/users/scook/proj/DE_Bay/run07/subtidal/de_bay_run07_subtidal.ncml.html

To access the data in xarray we use the OPeNDAP Data URL, and when I just tried, the above python code is still working.

[philippjfr]

Ah, thanks and sorry about the false alarm, I tried just downloading it so I could speed the process up.

[philippjfr]

I think I've finally got a real fix here, my previous approach was very obviously wrong after looking at it again. Will try to get 1.13.4 out asap.

[JessicaS11]

Any update on when we can expect 1.13.4? I recently started using Holoviews and have been liking it so far, but have been having to force restart my kernel frequently due to this issue. I was excited to learn that it was already a known bug that has a fix coming in the next release!

[philippjfr]

I'm currently trying to fix an issue building the documentation for HoloViews. Otherwise nothing is holding up the release now so I'm hoping today or tomorrow is the date.