Bokeh server support #959

philippjfr · 2016-10-31T01:36:24Z

Adds support for deploying any holoviews object as a bokeh server app.

BokehRenderer initializes server document
Initial support for handling stream events

To do:

Bokeh widget implementation to support DynamicMaps and HoloMaps with standard streams
Ensure that bokeh stream callbacks relying on cb_data and cb_obj work correctly.
Add example apps.

Optional/Future:

Provide easy way to tie a stream to any bokeh widget.
Support bokeh widgets in a notebook context.

Example app:

import numpy as np
import holoviews as hv
import holoviews.plotting.bokeh
from holoviews.operation.datashader import datashade

hv.Store.current_backend = 'bokeh'
renderer = hv.Store.renderers['bokeh'].instance(mode='server', holomap='server')
points = hv.HoloMap({i: hv.Points(np.random.multivariate_normal((0,0), [[0.1, 0.1], [0.1, 1.0]], (1000000,))) for i in range(10)})
datashaded = datashade(points, x_sampling=0.01, y_sampling=0.01)
doc, _ = renderer(datashaded)
doc.title = 'HoloViews Datashade'

jlstevens · 2017-02-03T17:00:58Z

Nice to have the examples!

That said, I feel they belong in contrib which is an entire repo dedicated to various examples. I would put them in the scripts directory here.

philippjfr · 2017-02-03T17:03:28Z

I do like the idea of contrib, but until it's been integrated into our doc building process and referenced throughout our documentation it just seems like no one will ever find it.

Edit: I do think it's the right place for these examples.

jlstevens · 2017-02-03T17:15:15Z

Given that we want to split out the bokeh backend eventually and this is definitely bokeh related, it should be in plotting.bokeh.examples if you don't want it in contrib.

philippjfr · 2017-02-03T21:23:41Z

it should be in plotting.bokeh.examples if you don't want it in contrib.

I did think contrib was the right place, but plotting.bokeh.examples might actually be even better if we are going to split it out.

jlstevens · 2017-02-03T21:46:20Z

I think it would be nice to keep all the examples together in contrib as it would show the breadth of what you can do with HoloViews. There is no restriction in contrib on backends or any other libraries necessary to make the examples work.

For this reason, I think we should move these examples to contrib right now - the bokeh backend repo could always link to the contrib examples to ensure people can discover them.

philippjfr · 2017-03-26T12:31:39Z

@jlstevens, @jbednar I've rebased this PR with the latest events PR and it's working as I want it to. The question is now about scope, I'll happily move the current examples to contrib, but my main question is about bokeh widgets. As of right now this PR ships with a small class that emulates our regular Dimension based widgets, and you can easily connect streams to other widgets and use the regular bokeh API to compose your new set of widgets into a layout with the plot(s).

Should we ship a basic widget class which generates widgets for HoloMap/DynamicMap key dimensions? You can see what that looks like in the gif example above. The implementation is fairly straightforward (~130 LOC) and means you can directly deploy standard HoloMaps/DynamicMaps you used and defined in a notebook without manually having to hook up widgets for key dimensions.

jbednar · 2017-03-26T14:19:03Z

That all sounds good, but is there a reason not to just use Bokeh widgets all the time when we are using the Bokeh backend, even in the notebook? That seems like the way to have seamless interoperability with bokeh in its various possible usages.

philippjfr · 2017-03-26T14:26:43Z

If we had already done the widget refactor and implemented a general widget/comm manager as we are planning to do I'd say yes. However for the time being I'd have to come up with some ad hoc solution which I'm strongly against. So my answer is that once we have that we can easily retrofit this widget implementation to support it.

In the long run we should find an easy and general way to map both dimensions and streams to specific widgets.

philippjfr · 2017-03-26T17:00:23Z

Waiting on bokeh/bokeh#6062 so that the cb_obj can be handled the same for the js_on_event and on_event cases.

I've also implemented an event queue on the python end which works fairly well. Once the PR has been merged I'll have to extend that implementation to handle different event types as I did on the JS end.

philippjfr · 2017-04-06T13:20:34Z

@jlstevens This is now ready for review. All streams now work for both bokeh server and in the notebook. I'm happy to move the example apps elsewhere though.

jlstevens · 2017-04-07T00:03:18Z

holoviews/plotting/bokeh/widgets.py

+        if self.plot.renderer.mode == 'default':
+            self.attach_callbacks()
+        self.state = self.init_layout()
+        self._event_queue = []


I would still rename this from _event_queue to just queue.

Ah, this is for the widgets now.. so it might be ok.

Although maybe better for consistency? I'll rename.

This is in widgets now so my original comment was addressed. Might be worth renaming here too - up to you.

jlstevens · 2017-04-07T00:18:32Z

holoviews/plotting/bokeh/callbacks.py

+    def _init_plot_handles(self):
+        """
+        Find all requested plotting handles and cache them along
+        with the IDs of the models callbacks will be attached to.


Either an an extra word here or words missing. Either '... along with the IDs the model callbacks will be attached to' or '...along with the IDs of the model callbacks that they will be attached to'.

Neither, it's missing a the:

along with the IDs of the models the callbacks will be attached to.

That works too. I knew something was wrong!

jlstevens · 2017-04-07T00:21:34Z

holoviews/plotting/bokeh/callbacks.py

+    def process_on_change(self):
+        if not self._queue:
+            return
+        self._queue = []


How come the _queue isn't used for process_on_change?

Because process_on_change just looks up the model values directly, that avoids having separate callbacks for each model change, which would be required otherwise.

jlstevens · 2017-04-07T00:22:07Z

holoviews/plotting/bokeh/callbacks.py

+    the data sent by the callback before it is passed to the streams.
+    """
+
+    def initialize(self):


Much better! :-)

jlstevens · 2017-04-07T00:24:39Z

holoviews/plotting/bokeh/renderer.py

+
+    mode = param.ObjectSelector(default='default',
+                                objects=['default', 'server'], doc="""
+         Whether to render the DynamicMap in regular or server mode. """)


Is it only rendering DynamicMaps? Might want a bit more to say what server mode is about ...

Hmm, not sure why I wrote that. It handles anything.

jlstevens · 2017-04-07T00:25:39Z

holoviews/plotting/bokeh/widgets.py

+
+
+    @classmethod
+    def create_widget(self, dim, holomap=None):


Much nicer as a classmethod! Good for testing...

jlstevens · 2017-04-07T00:27:43Z

Fantastic!

Other than one docstring fix I am now very happy with this PR. The functionality is great and the code is now structured in a much more sensible way.

I'll merge once that is done.

jlstevens · 2017-04-07T00:35:04Z

All done! Merging.

philippjfr added the status: WIP label Oct 31, 2016

philippjfr changed the title ~~Bokeh server~~ Bokeh server support Oct 31, 2016

philippjfr force-pushed the bokeh_server branch from 6cba793 to 2d80908 Compare November 1, 2016 01:16

philippjfr added this to the Major feature request milestone Nov 16, 2016

philippjfr mentioned this pull request Nov 29, 2016

ENH: Support for using bokeh through bokeh server #694

Closed

philippjfr mentioned this pull request Jan 25, 2017

Extend InteractiveImage to work with bokeh server holoviz/datashader#147

Closed

philippjfr force-pushed the bokeh_server branch from 2d80908 to c3151cb Compare February 3, 2017 14:08

jbednar mentioned this pull request Feb 7, 2017

InteractiveImage callbacks to Flask app holoviz/datashader#279

Closed

philippjfr mentioned this pull request Feb 25, 2017

Documentation for non-Ipython usage too sparse #1136

Open

philippjfr modified the milestones: v1.7.0, Major feature request Mar 15, 2017

philippjfr self-assigned this Mar 16, 2017

philippjfr force-pushed the bokeh_server branch from 0cd9548 to 16a6338 Compare March 26, 2017 11:52

philippjfr added tag: backend: bokeh type: feature A major new feature and removed status: WIP labels Mar 26, 2017

philippjfr force-pushed the bokeh_server branch from 8ad43db to 0f4e00f Compare March 26, 2017 17:20

philippjfr force-pushed the bokeh_server branch from 9f9bf0d to 833fd87 Compare April 6, 2017 13:18

philippjfr added the status: ready label Apr 6, 2017

philippjfr force-pushed the bokeh_server branch from 833fd87 to cc7a700 Compare April 6, 2017 13:22

philippjfr added 7 commits April 6, 2017 22:31

Moved bokeh server example apps

dbef691

Completely refactored bokeh Callbacks

2e9ca71

Made callback utilities into classmethods

d7f5e45

Minor cleanup on bokeh Callbacks

6ab8861

Added tests for bokeh Callbacks

82074b8

Small fix for bokeh ServerCallback on_change events

a8a10a5

Allow supplying Document to BokehRenderer

3342b0a

philippjfr force-pushed the bokeh_server branch from fba7bb4 to 3342b0a Compare April 6, 2017 21:31

philippjfr added 6 commits April 6, 2017 23:18

Simplified bokeh Callback initialization

c65dcbb

Moved bokeh server widget handling onto BokehRenderer

a2fcb0c

Factored out class method to create bokeh widgets

b0e1f52

Small fixes and improvements for bokeh widgets

d9fe1b7

Added tests for BokehServerWidgets

f0c31c6

Fixed unreferenced variable bugs

171e9ca

philippjfr force-pushed the bokeh_server branch from 4b63765 to 171e9ca Compare April 6, 2017 23:59

jlstevens reviewed Apr 7, 2017

View reviewed changes

Various python3 fixes

4e8073a

jlstevens reviewed Apr 7, 2017

View reviewed changes

Improved docstrings for bokeh server features

957b96b

jlstevens merged commit 3503bb0 into master Apr 7, 2017

jlstevens mentioned this pull request Apr 10, 2017

SelectionWidget should follow structure of BokehServerWidgets #1258

Closed

philippjfr deleted the bokeh_server branch April 11, 2017 12:19

stevievb mentioned this pull request Jun 13, 2017

datashader on bokeh server #1536

Closed

oarcher mentioned this pull request Apr 19, 2019

no stacktrace printed in bokeh serve mode after import datashader #3638

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Bokeh server support #959

Bokeh server support #959

philippjfr commented Oct 31, 2016 •

edited

jlstevens commented Feb 3, 2017

philippjfr commented Feb 3, 2017 •

edited

jlstevens commented Feb 3, 2017

philippjfr commented Feb 3, 2017

jlstevens commented Feb 3, 2017

philippjfr commented Mar 26, 2017 •

edited

jbednar commented Mar 26, 2017

philippjfr commented Mar 26, 2017

philippjfr commented Mar 26, 2017

philippjfr commented Apr 6, 2017

jlstevens Apr 7, 2017 •

edited

jlstevens Apr 7, 2017

philippjfr Apr 7, 2017

jlstevens Apr 7, 2017

jlstevens Apr 7, 2017

philippjfr Apr 7, 2017

jlstevens Apr 7, 2017

jlstevens Apr 7, 2017

philippjfr Apr 7, 2017

jlstevens Apr 7, 2017

jlstevens Apr 7, 2017

philippjfr Apr 7, 2017

jlstevens Apr 7, 2017

jlstevens commented Apr 7, 2017

jlstevens commented Apr 7, 2017

Bokeh server support #959

Bokeh server support #959

Conversation

philippjfr commented Oct 31, 2016 • edited

jlstevens commented Feb 3, 2017

philippjfr commented Feb 3, 2017 • edited

jlstevens commented Feb 3, 2017

philippjfr commented Feb 3, 2017

jlstevens commented Feb 3, 2017

philippjfr commented Mar 26, 2017 • edited

jbednar commented Mar 26, 2017

philippjfr commented Mar 26, 2017

philippjfr commented Mar 26, 2017

philippjfr commented Apr 6, 2017

jlstevens Apr 7, 2017 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jlstevens commented Apr 7, 2017

jlstevens commented Apr 7, 2017

philippjfr commented Oct 31, 2016 •

edited

philippjfr commented Feb 3, 2017 •

edited

philippjfr commented Mar 26, 2017 •

edited

jlstevens Apr 7, 2017 •

edited