Creating GlyphRenderers for modifying the legend - python-3.x

Question

I want to create a bokeh application that can filter points based on some attribute. Here is a very simple code example for my use case that filters points on the plot using checkboxes.
from bokeh.plotting import ColumnDataSource, figure, curdoc
import bokeh.models as bmo
from bokeh.layouts import row
import numpy as np
def update_filter(selected_colors):
keep_indices = []
for i, color in enumerate(cds.data['color']):
if color2idx[color] in selected_colors:
keep_indices.append(i)
view.filters[0] = bmo.IndexFilter(keep_indices)
cds = ColumnDataSource(data=dict(
x=np.random.rand(10),
y=np.random.rand(10),
color=['red', 'green', 'blue', 'red', 'green',
'blue', 'red', 'green', 'blue', 'red'])
)
view = bmo.CDSView(source=cds, filters=[bmo.IndexFilter(np.arange(10))])
checkboxes = bmo.CheckboxGroup(labels=['red', 'green', 'blue'], active=[0, 1, 2])
color2idx = {'red': 0, 'green': 1, 'blue': 2}
checkboxes.on_change('active', lambda attr, old_val, new_val: update_filter(new_val))
fig = figure(plot_width=400, plot_height=400, title='Visualize')
fig.circle(x='x', y='y', fill_color='color', size=10, source=cds, view=view, legend_field='color')
curdoc().add_root(row(checkboxes, fig))
curdoc().title = 'Plot'
It works well, however, when I filter points out by de-selecting one of the checkboxes, the legend becomes erroneous.
Below is a screenshot when all the colors are selected:
And this is a screenshot when one of the colors is de-selected:
As it can be seen, the legend for "green" became red in color when the checkbox for "green" was de-selected.
I found that legends do not work properly with CDSView and it is still an unsolved issue: https://github.com/bokeh/bokeh/issues/8010
So, I wrote the function below that would modify the legend so that it is not erroneous.
def update_legend():
# Find the indices in the CDS that are visible
filters = view.filters
visible_indices = set(list(range(len(cds.data['x']))))
for filter in filters:
visible_indices = visible_indices & set(filter.indices)
# Get a list of visible colors
visible_colors = set([cds.data['color'][i] for i in visible_indices])
# Create a dummy figure to obtain renderers
dummy_figure = figure(plot_width=0, plot_height=0, title='')
legend_items = []
# Does not work
for color in visible_colors:
renderer = dummy_figure.circle(x=[0], y=[0], fill_color=color, size=10)
legend_items.append(bmo.LegendItem(label=color, renderers=[renderer]))
fig.legend[0].items = legend_items
And added another event callback for the checkbox group:
checkboxes.on_change('active', lambda attr, old_val, new_val: update_legend())
When I did the above, the labels in the legend were corrected but now the glyphs are not rendered in the legend. Below is a screenshot of the same:
What am I doing wrong? How should I create a GlyphRenderer for the legend such that the issue gets resolved?

Answer 1

This works for Bokeh v2.1.1. In addition to your original code you can also click on a legend item to show/hide the circles.
from bokeh.plotting import ColumnDataSource, figure, curdoc
from bokeh.models import CheckboxGroup, Row, CDSView, IndexFilter
import numpy as np
colors = ['red', 'green', 'blue']
cds = ColumnDataSource(dict(x=np.random.rand(10),
y=np.random.rand(10),
color=['red', 'green', 'blue', 'red', 'green', 'blue', 'red', 'green', 'blue', 'red']))
def update_filter(selected_colors):
for i in range(len(colors)):
renderers[i].visible = True if i in selected_colors else False
checkboxes = CheckboxGroup(labels=colors, active=[0, 1, 2], width = 50)
checkboxes.on_change('active', lambda attr, old_val, new_val: update_filter(new_val))
fig = figure(plot_width=400, plot_height=400, title='Visualize')
views = [CDSView(source=cds, filters=[IndexFilter([i for i, x in enumerate(cds.data['color']) if x == color])]) for color in colors]
renderers = [fig.circle(x='x', y='y', fill_color='color', size=10, source=cds, view=views[i], legend=color) for i,color in enumerate(colors)]
fig.legend.click_policy = 'hide'
curdoc().add_root(Row(checkboxes, fig))
curdoc().title = 'Plot'
Result: