Jupyter notebooks are computable documents often used for exploratory work, data analysis, teaching, and demonstration. A notebook is a series of input cells that can be individually executed to display their output immediately after the cell. In addition to Classic Notebooks there are also notebooks for the newer JupyterLab project. Bokeh can embed both standalone and Bokeh server content with either.
Standalone Bokeh content (i.e. that does not use a Bokeh server) can be embedded directly in classic Jupyter notebooks as well as in JupyterLab.
To display Bokeh plots inline in a classic Jupyter notebooks, use the output_notebook() function from bokeh.io instead of (or in addition to) the output_file() function we have seen previously. No other modifications are required. When show() is called, the plot will be displayed inline in the next notebook output cell. You can see a Jupyter screenshot below:
output_notebook()
output_file()
show()
Multiple plots can be displayed in a single notebook output cell by calling show() multiple times in the input cell. The plots will be displayed in order.
In order to embed Bokeh plots inside of JupyterLab, you need to install the two JupyterLab extensions. First install jupyterlab-manager by running the following command:
jupyter labextension install @jupyter-widgets/jupyterlab-manager
And then similarly install the jupyter_bokeh extension:
jupyter labextension install @bokeh/jupyter_bokeh
Once this is installed, usage is the same as with classic notebooks above.
It is also possible to embed full Bokeh server applications, that can connect plot events, and Bokeh’s built-in widgets, directly to Python callback code. See Running a Bokeh Server for general information about Bokeh server applications, and the following notebook for a complete example of a Bokeh application embedded in a Jupyter notebook:
examples/howto/server_embed/notebook_embed.ipynb
In order to embed Bokeh server applications when running notebooks from your own JupyterHub instance, some additional steps are necessary to enable network connectivity between the client browser and the Bokeh server running in the JupyterLab cell. This is because your browser needs to connect to the port the Bokeh server is listening on, but JupyterHub is acting as a reverse proxy between your browser and your JupyterLab container. Follow all the JupyterLab instructions above, then continue with the following steps below.
First, you must install the nbserverproxy server extension This can be done by running the command:
nbserverproxy
pip install nbserverproxy && jupyter serverextension enable --py nbserverproxy
Second, you must define a function to help create the URL that the browser uses to connect to the Bokeh server. This will be passed into show() in the final step. A reference implementation is provided here, although you must either modify it or define the environment variable EXTERNAL_URL to the URL of your JupyterHub installation. By default, JupyterHub will set JUPYTERHUB_SERVICE_PREFIX.
EXTERNAL_URL
JUPYTERHUB_SERVICE_PREFIX
def remote_jupyter_proxy_url(port): """ Callable to configure Bokeh's show method when a proxy must be configured. If port is None we're asking about the URL for the origin header. """ base_url = os.environ['EXTERNAL_URL'] host = urllib.parse.urlparse(base_url).netloc # If port is None we're asking for the URL origin # so return the public hostname. if port is None: return host service_url_path = os.environ['JUPYTERHUB_SERVICE_PREFIX'] proxy_url_path = 'proxy/%d' % port user_url = urllib.parse.urljoin(base_url, service_url_path) full_url = urllib.parse.urljoin(user_url, proxy_url_path) return full_url
Finally, you can pass the function you defined in step 2 to show() as the notebook_url keyword argument, which Bokeh will call while setting up the server and creating the URL for loading the graph:
show(obj, notebook_url=remote_jupyter_proxy_url)
At this point, the Bokeh graph should load and execute python callbacks defined in your JupyterLab environment.
Depending on the version of the Notebook in use, it may be necessary to “trust” the notebook in order for Bokeh plots to re-render when the notebook is closed and subsequently re-opened. The “Trust Notebook” option is typically located under the “File” menu:
It is possible to use the Jupyter notebook in conjunction with Reveal.js to generate slideshows from notebook cell content. It is also possible to include standalone (i.e. non-server) Bokeh plots in such sideshows, however some steps must be followed for output to correctly display. Primarily: the cell containing output_notebook must be not be skipped.
output_notebook
The rendered cell output of the output_notebook call is responsible for making sure the BokehJS library is loaded. Without that, Bokeh plots cannot function. If this cell type is marked “skip” then BokehJS will not be loaded, and Bokeh plots will not display. An alternative, if you wish to hide this cell, is to mark it as the “notes” slide type.
It is possible to update a previously shown plot in-place. When the argument notebook_handle=True is passed to show() then a handle object is returned. This handle object can be used with the push_notebook() function to update the plot with any recent changes to plots properties, data source values, etc. This notebook handle functionality is only supported in classic Jupyter notebooks and is not implemented in JupyterLab or Zeppelin yet.
notebook_handle=True
push_notebook()
The following screenshots walk through the basic usage of notebook handles.
First, import standard functions, as well as push_notebook():
Next, create some plots, and make sure to pass notebook_handle=True to show():
Looking at the handle, see that it is associated with the output cell for In[2] that was just displayed:
In[2]
Now, update any properties of the plot, then call push_notebook() with the handle:
After doing so, note that the earlier output cell for In[2] has changed (without being re-executed)
More detailed demonstrations of using notebook handles can be found in the following example notebooks:
examples/howto/notebook_comms/Basic Usage.ipynb
examples/howto/notebook_comms/Continuous Updating.ipynb
examples/howto/notebook_comms/Jupyter Interactors.ipynb
examples/howto/notebook_comms/Numba Image Example.ipynb
It is possible to drive updates to Bokeh plots using Jupyter notebook widgets, known as interactors. The key doing this is the push_notebook() function described above. Typically it is called in the update callback for the interactors, to update the plot from widget values. A screenshot of the examples/howto/notebook_comms/Jupyter Interactors.ipynb example notebook is shown below:
Many more examples using Jupyter Notebook can be found in the bokeh-notebook repository. First clone the repository locally:
git clone https://github.com/bokeh/bokeh-notebooks.git
Then launch Jupyter Notebook in your web browser. Alternatively, live notebooks that can be run immediately online are hosted by Binder.
Additionally, there are some notebooks under examples in the main Bokeh repo:
categorical data
hover callback
linked panning
range update callback
embed server in notebook
US marriages and divorces interactive
color scatterplot
glyphs
Notebook comms examples:
basic usage
continuous updating
Jupyter interactors
Numba image example
In the previous section we learnt how to use Bokeh in JupyterLab and classical notebook environments. Suppose we would like to do the opposite and take advantage of the vibrant Jupyter ecosystem, in particular IPyWidgets, in a Bokeh application, outside the confines of those environments. This can be achieved with help from ipywidgets_bokeh extension to Bokeh:
$ conda install -c bokeh ipywidgets_bokeh
or
$ pip install ipywidgets_bokeh
Then you can use an IPyWidget in Bokeh, by simply wrapping it in IPyWidget model and adding the wrapper to a document or including it in a layout. Given that this is run outside Jupyter, there is no need for installing and/or enabling any extensions.
IPyWidget
Suppose we would like to create an application with a single Jupyter slider and log its value to the console, as the slider is manipulated. We start by constructing the widget and configuring an observer, the same as we would do in Jupyter:
from ipywidgets import FloatSlider angle = FloatSlider(min=0, max=360, value=0, step=1, description="Angle") def on_change(change): print(f"angle={change['new']} deg") angle.observe(on_change, names="value")
To integrate the widget with Bokeh, we have to wrap it in IPyWidget:
from ipywidgets_bokeh import IPyWidget ipywidget = IPyWidget(widget=angle)
Then we add the wrapper to a Bokeh document:
from bokeh.plotting import curdoc doc = curdoc() doc.add_root(ipywidget)
To run this, assuming the code is saved under ipy_slider.py, we issue bokeh serve ipy_slider.py (see Running a Bokeh Server for details). The application is available at http://localhost:5006/ipy_slider.
ipy_slider.py
bokeh serve ipy_slider.py
From here, one can create more complex layouts and include advanced widgets, like ipyleaflet, ipyvolume, etc.