Update datashader map without recreating figure/tiles


I’m working on a dashboard to showcase radar data with some controls on the sidebar. Current dashboard look like this:

This map is currently plot using this function:

@pn.depends(var_select.param.value, angle_select.param.value, player.param.value)
def graph(selected_var, angle, timestep): 
    # get timestamp
    timestamp = ds['time'][timestep].astype(int)
    ns = 1e-9
    dt = datetime.utcfromtimestamp(timestamp * ns)
    ts = dt.strftime("%Y/%m/%d %H:%M:%S")

    data = ds[selected_var].sel(scan_angle=angle).isel(time=timestep).load() 
    if selected_var in ['Reflectivity', 'CorrectedReflectivity','DifferentialReflectivity', 'CorrectedDifferentialReflectivity']:
        color_map = NWSReflectivity
    elif selected_var in ['Velocity']:
        color_map = NWSVelocity
        color_map = 'gist_ncar'
    data = hv.Dataset(data, vdims=selected_var)
    graph = data.to(gv.Image, ["lon", "lat"], dynamic=True)
    tiles = gv.tile_sources.StamenTonerBackground().apply.opts()
    # site
    ntc = gv.Text(-73.82, 40.7, 'National Tennis Center').opts(text_font_style='bold')
    site = gv.Points([(-73.8492017,40.7498823)]).opts(color='red', line_color='black', line_width=0.5, size=5)
    site_shapely = Point(-73.8492017,40.7498823)
    first_circle = gv.Shape(site_shapely.buffer(0.0573)).opts(line_color='red', line_dash='dashed', fill_alpha=0,line_width=2, alpha=0.5) #3mi
    second_circle = gv.Shape(site_shapely.buffer(0.0955)).opts(line_color='blue', line_dash='dashed',  fill_alpha=0, line_width=2, alpha=0.5) #5mi
    third_circle = gv.Shape(site_shapely.buffer(0.191)).opts(line_color='green', line_dash='dashed',  fill_alpha=0, line_width=2, alpha=0.5) #10mi
    cirle_anno1 = gv.Text(-73.8492017, 40.807, '3mi').opts(text_alpha=0.6, text_font_size='12px')
    cirle_anno2 = gv.Text(-73.8492017, 40.846, '5mi').opts(text_alpha=0.6, text_font_size='12px')
    cirle_anno3 = gv.Text(-73.8492017, 40.941, '10mi').opts(text_alpha=0.6, text_font_size='12px')
    graph = (
        rasterize(graph).opts(title=ts, height=550, width=800, colorbar=True, cnorm='eq_hist', cmap=color_map, alpha=0.5, tools=['hover'])
    ) * tiles * site * ntc * first_circle * second_circle * third_circle * cirle_anno1 * cirle_anno2 * cirle_anno3
    return graph

And to serve in the dashboard:

dark_material = pn.template.MaterialTemplate(title='US Open Skyler', theme='dark')


The issue I have is, in this current approach, every parameter change, the graph function got called once, and the ds.sel got called once, and generate a new map. I saw in the Xarray quadmesh — GeoViews 1.9.1+g33876c8-dirty documentation example, this build-in slider when converting a xarray.Dataset to a geoviews.Dataset, and the transition between timestamp is much smoother without recreating the map.


  1. How to use and customize the built-in controls if a Dataset is 3D or 4D, and served them in sidebar?
  2. If not, how to frame the graph function to only update the data, without recreating a new map whenever called?

Thank you very much!

I think you need a separate load function and a plot function, then wrap a DynamicMap around your plot function.Tips and Tricks — pYdeas 0.0.0 documentation

Right; you’ll need to separate the various bits of computation that are currently all lumped together if you want just the updated data to refresh. See examples at https://anaconda.org/jbednar/dashboard_barewidgets/notebook , but also note that nowadays I strongly, strongly recommend using pn.bind rather than @pn.depends, because it lets code be organized in a much more natural way (separating GUI logic from analytics calculations).

1 Like

Thanks for reply! I managed learn something useful from your replies and the two great examples.

I managed to wrap the hv.DynamicMap outside pn.bind like:

import holoviews as hv
skyler = SkylerApp(ds)

test = hv.DynamicMap(pn.bind(skyler.make_map_bind, skyler.variable.param.value ,skyler.angle.param.value, skyler.player.param.value))
        pn.Column(skyler.variable, skyler.angle, skyler.player), test

the make_map_bind looks like this:

    def make_map_bind(self, var_selected, scan_angle, timestep):
        ### TODO how to handle the dependency
        ### https://panel.holoviz.org/user_guide/Param.html
        timestamp = self._ds['time'][timestep].astype(int)
        ns = 1e-9
        dt = datetime.utcfromtimestamp(timestamp * ns)
        ts = dt.strftime("%Y/%m/%d %H:%M:%S")
        # Need to solve this part
        data = self._ds[var_selected].sel(scan_angle=scan_angle).isel(time=timestep).load() 
        NWSVelocity = [
        NWSReflectivity = [
        if var_selected in ['Reflectivity', 'CorrectedReflectivity','DifferentialReflectivity', 'CorrectedDifferentialReflectivity']:
            color_map = NWSReflectivity
        elif var_selected in ['Velocity']:
            color_map = NWSVelocity
            color_map = 'gist_ncar'
        data = hv.Dataset(data, vdims= var_selected)
        graph = data.to(gv.Image, ["lon", "lat"]).opts(title=ts, height=550, width=800, colorbar=True, cnorm='eq_hist', cmap=color_map, alpha=0.5, tools=['hover'])
#         graph = (
#             rasterize(hv.DynamicMap(graph)).opts(title=ts, height=550, width=800, colorbar=True, cnorm='eq_hist', cmap=color_map, alpha=0.5, tools=['hover'])
#         ) 
        return graph * self._base

However, using rasterize, the plot seems to refresh itself much faster (but with blinking, and the refresh the base as well.) Wrapping hv.DynamicMap outside pn.bind does keep the map tile intact, but the data refresh themselves much slower. Wrapping DyamicMap outside rasterize is not allowed due to nested DynamicMap. Any tips?