Create a grid of linked interactive bokeh-datashader plots

I would like to create a grid of bokeh-datashader plots using something like:

plots = []
plots.append(InteractiveImage(p1, image_callback1))
plots.append(InteractiveImage(p2, image_callback2))
plots.append(InteractiveImage(p2, image_callback3))
bokeh.layouts.gridplot([[plots[0], plots[1], plots[2]]])

while also linking the plot ranges using something like this

for p in plots:
    p.x_range = plots[0].x_range
    p.y_range = plots[0].y_range

so that adjusting the range of one of the plots also adjusts the range of all the plots, while updating the datashader displays. I realize what I have does not work, because InteractiveImage does not return a bokeh plot object. Is there currently a way to do this?

[jbednar]

HoloVIews 1.7 has been released, and can be installed using conda install -c ioam holoviews. This version includes full support for datashader, as outlined in the greatly expanded HoloViews Datashader notebook. You should be able to run that notebook and see how to combine multiple Datashader plots into any arrangement, e.g. (q1+q2+q3+q4+q5+q6).cols(3). Note that datashader itself has not yet been released, so until it is you’ll need the git master version.

[jbednar]

It might be possible to modify InteractiveImage to do this, but instead I think you ought to be using HoloViews for anything but a single, isolated datashader plot. InteractiveImage is a tiny piece that we originally took out of HoloViews to show off how datashader works, but for anything ambitious, I’d recommend just using HoloViews itself.

The HoloViews API is formulated at a level one step higher than Bokeh’s, which makes it vastly simpler to do things like dynamically updating an underlying plot, as InteractiveImage needs to do, while also embedding the plot into a larger layout. E.g. combining two interactive datashader-based Bokeh plots X and Y with linked axes in HoloViews is normally just “X+Y”.

We are gearing up for a release of HoloViews that has documentation for how to use it with datashader, but meanwhile there are examples in this HoloViews Datashader notebook. Code for doing what you outline above should be one or two lines, in HoloViews. Basically, use NdLayout if your “Q” values can be interpreted as a dimension (values along some axis), and otherwise use Layout (which is what the “+” operator constructs).