Holoviews linked_selection on 2 images sharing same coordinates and a scatter

[oarcher]

I’ve got 2 datasets sharing the sames coordinates. I’m able to plot both images and the corresponding flattened scatter :

What I want is a working ‘lasso_select’, that will highlight selected points in all plots. I’ve tried to use ‘linked_selection’ from Linked_Brushing, without success.

I’m using
bokeh 2.0.2 and holoviews 1.13.2

Here is the code that show the problem:

! (cd /tmp ; wget -N ftp://ftp.ifremer.fr/ifremer/cersat/datarmor/project/sarwing/users/oarcher/tmp/rs2.nc)
! (cd /tmp ; wget -N ftp://ftp.ifremer.fr/ifremer/cersat/datarmor/project/sarwing/users/oarcher/tmp/s1.nc)

import holoviews as hv
hv.extension('bokeh')
import xarray as xr

# read dataset. They have same coordinates names and value (ie same grid)
ds_s1 = xr.open_dataset('/tmp/s1.nc')
ds_rs2 = xr.open_dataset('/tmp/rs2.nc')

# get wind_speed dataarray
da_s1 = ds_s1['wind_speed']
da_rs2 = ds_rs2['wind_speed']

# two Image
plots = [hv.Image(da).opts(cmap='jet',tools=['hover', 'lasso_select']) for da in [da_rs2,da_s1]]

# add scatter plot, after flatenning ('x','y')
da_rs2_lin = da_rs2.stack(rs2=('x', 'y'))
da_s1_lin  = da_s1.stack(s1=('x', 'y'))
plots.append(hv.Scatter((da_rs2_lin,da_s1_lin),kdims=['rs2'],vdims=['s1']))

# not linked: it works
layout=hv.Layout(plots)
#layout.opts(hv.opts.Image(tools=['hover', 'tap', 'lasso_select']),hv.opts.Scatter(tools=['hover', 'tap', 'lasso_select']))

# no difference using link_selection
from holoviews.selection import link_selections
link_selections(layout,index_cols=['x','y']).opts(hv.opts.Image(tools=['hover', 'tap', 'lasso_select']),hv.opts.Scatter(tools=['hover', 'tap', 'lasso_select']))

What I am missing ?

[philippjfr]

Apart from the fact that that lasso select is currently only supported on master and will be part of the 1.13.3 release, there is one other issue with the way your example is structured. In particular HoloViews cannot figure out that your data has x/y dimensions on the Scatter since it interprets the two data arrays as simple arrays. The easiest way of ensuring that plots can be linked is usually by starting out with a single hv.Dataset and then creating all your plots from that. So here we manually construct ds from the x and y coordinates and the two DataArrays and then construct all of the plots from that:

import holoviews as hv
import xarray as xr

from holoviews.selection import link_selections

hv.extension('bokeh')

# read dataset. They have same coordinates names and value (ie same grid)
ds_s1 = xr.open_dataset('/tmp/s1.nc')
ds_rs2 = xr.open_dataset('/tmp/rs2.nc')

# get wind_speed dataarray
da_s1 = ds_s1['wind_speed']
da_rs2 = ds_rs2['wind_speed']

ds = hv.Dataset((da_rs2.x, da_rs2.y, da_rs2.values, da_s1.values), ['x', 'y'], ['rs2', 's1'])

# two Image
plots = [hv.Image(ds, vdims=vdim).opts(cmap='jet',tools=['hover', 'lasso_select']) for vdim in ds.vdims]
plots.append(hv.Scatter(ds, 'rs2', 's1'))

layout=hv.Layout(plots)

link_selections(layout).opts(hv.opts.Image(tools=['hover', 'tap', 'lasso_select']),hv.opts.Scatter(tools=['hover', 'tap', 'lasso_select']))

[oarcher]

That’s exactly what I need.
Thanks !