View vs. Viewable with displaying widget

Question

I am putting together an interactive dashboard using the pyviz ecosystem. One feature of the dashboard is that the underlying data may change based on a widget selector. Below is an example code showing the issue I have with getting the time widget slider to appear:

Package Versions:
panel: 0.5.1
param: 1.9.0
holoviews: 1.12.3
geoviews: 1.6.2

Example:

import xarray as xr
import panel as pn
import numpy as np
import param as pm
import holoviews as hv
import geoviews as gv
from matplotlib import cm
import geoviews.tile_sources as gts
from holoviews.operation.datashader import rasterize
from collections import OrderedDict as odict
from holoviews import opts
renderer = hv.renderer('bokeh')
pn.extension()

dset = xr.DataArray(np.random.random((100,100,100)),coords={'X':np.arange(100),'Y':np.arange(100),'T':np.arange(100)},dims=['X','Y','T']).to_dataset(name='test')
dset = gv.Dataset(dset, ['X', 'Y', 'T'], 'test').to(gv.QuadMesh, groupby='T').opts(cmap='viridis', colorbar=True, show_frame=False)

fields = odict([('test','test')])#odict([(v.get('label',k),k) for k,v in source.metadata['fields'].items()])
aggfns = odict([(f.capitalize(),f) for f in ['mean','std','min','max','Pixel Level']])#'count','sum','min','max','mean','var','std']])#,'None (Pixel Level)']])
cmaps  = odict([(n,cm.get_cmap(n)) for n in ['viridis','seismic','cool','PiYG']])
maps   = ['EsriImagery','EsriNatGeo', 'EsriTerrain', 'OSM']
bases  = odict([(name, gts.tile_sources[name].relabel(name)) for name in maps])
gopts  = hv.opts.WMTS(responsive=True, xaxis=None, yaxis=None, bgcolor='black', show_grid=False)


class Explorer_Test(pm.Parameterized):
    field = pm.Selector(fields)
    cmap = pm.Selector(cmaps)
    basemap = pm.Selector(bases)
    data_opacity = pm.Magnitude(1.00)
    map_opacity = pm.Magnitude(1.00)
    agg_fn_ = pm.Selector(aggfns,label='Aggregation**',default='mean')

    @pm.depends('field', 'agg_fn_')
    def aggregator(self):
        field = None if self.field == "counts" else self.field
        return self.agg_fn(field)

    @pm.depends('map_opacity', 'basemap')
    def tiles(self):
        return self.basemap.opts(gopts).opts(alpha=self.map_opacity)

    def viewable(self,**kwargs):
        rasterized = rasterize(dset, precompute=True).opts(colorbar=True, height=800, show_frame=False).apply.opts(cmap=self.param.cmap,alpha=self.param.data_opacity)
        return hv.DynamicMap(self.tiles)*rasterized

explorer_test = Explorer_Test(name="")

When I display the plot like:

panel = pn.Row(pn.Param(explorer_test.param, expand_button=False),explorer_test.viewable())
panel.servable()

The time widget appears:

Whereas:

panel = pn.Row(pn.Param(explorer_test.param, expand_button=False),explorer_test.viewable)
panel.servable()

In the first example, if I select an alternative dataset (based on a param.Selector widget - not shown in this example) it does not redraw the image. However, in the 2nd example, the image is redrawn, but I am missing the time slider.

UPDATE - Solution

Here is the workaround as per James' solutions (thanks!). This example includes changing the dataset and the variable (within each dataset) and the time parameter.

import xarray as xr
import panel as pn
import numpy as np
import param as pm
import holoviews as hv
import geoviews as gv
from holoviews.operation.datashader import rasterize
from collections import OrderedDict as odict
renderer = hv.renderer('bokeh')
pn.extension()

#Define Example Datasets
dset1 = xr.merge([xr.DataArray(np.random.random((50,50,50)),coords={'X':np.arange(50),'Y':np.arange(50),'T':np.arange(50)},dims=['X','Y','T']).to_dataset(name='var1'),
                  xr.DataArray(np.random.random((50,50,10))*.1,coords={'X':np.arange(50),'Y':np.arange(50),'T':np.arange(10)},dims=['X','Y','T']).to_dataset(name='var2')])
dset2 = xr.DataArray(np.random.random((50,50,20))*10,coords={'X':np.arange(50)/2.,'Y':np.arange(50)/3.,'T':np.arange(20)},dims=['X','Y','T']).to_dataset(name='var1')
data_dict = {'dset1':dset1,'dset2':dset2}                 

#Plot Datasets
class sel_dset_var():
    def dset1_var1():
        return rasterize(gv.Dataset(dset1.var1, ['X', 'Y', 'T'], 'test1').to(gv.QuadMesh, groupby='T')()).opts(cmap='viridis',colorbar=True, height=200, show_frame=False)
    def dset1_var2():
        return rasterize(gv.Dataset(dset1.var2, ['X', 'Y', 'T'], 'test1').to(gv.QuadMesh, groupby='T')()).opts(cmap='viridis',colorbar=True, height=200, show_frame=False)
    def dset2_var1():
        return rasterize(gv.Dataset(dset2.var1, ['X', 'Y', 'T'], 'test1').to(gv.QuadMesh, groupby='T')()).opts(cmap='viridis',colorbar=True, height=200, show_frame=False)

#Dashboard
class Explorer_Test(pm.Parameterized):
    dset = pm.Selector(odict([('Dataset1','dset1'),('Dataset2','dset2')]),default='dset1')
    varss = pm.Selector(list(dset1.data_vars),default=list(dset1.data_vars)[0])
    time1 = pm.Selector(dset1.var1.coords['T'].values,default=dset1.var1.coords['T'].values[0])

    @pm.depends('dset',watch=True)
    def update_var(self):
        self.param['varss'].objects = list(data_dict[self.dset].data_vars)
        self.param.set_param(varss=list(data_dict[self.dset].data_vars)[0])

    @pm.depends('dset',watch=True)
    def update_var(self):
        self.param['varss'].objects = list(data_dict[self.dset].data_vars)
        self.param.set_param(varss=list(data_dict[self.dset].data_vars)[0])

    def elem(self):
        return getattr(sel_dset_var,self.dset+'_'+self.varss)()

    @pm.depends('varss','dset',watch=True)
    def update_time(self):
        self.param['time1'].objects =data_dict[self.dset][self.varss].dropna(dim='T').coords['T'].values
        self.param.set_param(time1=data_dict[self.dset][self.varss].dropna(dim='T').coords['T'].values[0])

    def elem_yr(self):
        return getattr(self.elem(),'select')(T=self.time1)


    def viewable(self,**kwargs):
        return self.elem_yr

explorer_test = Explorer_Test(name="")
panel = pn.Row(pn.Param(explorer_test.param, expand_button=False),explorer_test.viewable())
panel.servable()

Cheers!

Answer 1

This code looks like it's derived from my http://datashader.org/dashboard.html example. In my example, the output from the viewable() method is already fully dynamic, and does not ever need to be regenerated, being already linked internally to all the widgets and controls that affect how it appears. Whereas if you pass viewable as a method name to Panel (rather than result of calling that method), you're asking Panel to call viewable() for you whenever it determines that the result from an initial call becomes stale. This simple re-run-the-method approach is appropriate for very simple cases of all-or-nothing computation, but not really useful here when the objects are already dynamic themselves and where specific controls are tied to specific aspects of the plot. (Why you also don't get a time widget in that case I'm not sure; it's not a recommended usage, but I would have thought it should still work in giving you a widget.)

Anyway, I don't think you should be trying to get the second case above to work, only the first one. And there the problem isn't the lack of the slider, it sounds like it's that you're trying to get the plot to be responsive to changes in your data source. Luckily, that case is already illustrated in the example in http://datashader.org/dashboard.html ; there rasterize dynamically wraps a method that returns the appropriate column of the data to show. You should be able to adapt that approach to make it dynamically reflect the state of some other widget that lets the user select the dataset.