Bokeh 0.12.9 Released


Hot on the heels of the last version, we are pleased to announce the release of Bokeh 0.12.9!

This update includes the following important new features:

  • Integration with JupyterLab
  • A much faster binary transport option for arrays

This release also contains several bugfixes and docs additions. For full details, see the CHANGELOG and Release Notes. Additionally, examples for each release are uploaded to CDN. Click here to download.

If you are using Anaconda, Bokeh can most easily be installed by executing the command conda install -c bokeh bokeh . Otherwise, use pip install bokeh.

First a quick note. You may have noticed that this announcement is not for 0.12.8 as expected. Last night, about 10 minutes after the release build finished, a showstopper issue was discovered. A “re-release” was attempted but was not possible. So here we are, early, at 0.12.9!

This problem was caught by the extremely astute and eager user @Pierretherenard. I very much appreciate their help late last night to resolve things, and also want stress: if you are an active Bokeh user, please always help us test by installing and trying out dev builds and release candidates. The more eyes there are on things, the more likely issues can be caught earlier, and that is better for everyone.

Now on to the fun bits!

JupyterLab Integration

In case you missed it, the first ever JupyterCon was recently held in New York. One of the big topics of discussion there was JupyerLab, which is the next generation of the Jupyter UI. JupyterLab is a robust and extensible framework that includes a new Notebook component, built from the ground up.

Luke Canavan added support for embedding Bokeh content in JupyterLab notebooks by creating a new jupyterlab_bokeh extension. You can install the extension by running

jupyter labextension install jupyterlab_bokeh

From your command line.

Here’s a screen cap of Bokeh+JLab in action:

Image of ridge plot in JupyterLab

The extension supports standalone plots and embedded Bokeh server applications. Updating plots via is not yet supported in JupyterLab, but will be soon. You can follow progress towards getting this feature working at issue #6619.

Efficient Binary Transport for Arrays

The Bokeh protocol has iterated and improved over the years. In the beginning, Bokeh’s approach was simply to “JSON all the things”. This was OK to start but quickly ran into performance issues when dealing with things like RGBA image data. Last year a Base64 encoding option was added for arrays, which offered a significant improvement. But the eventual goal has always been to offer a pure binary transport option for arrays when possible. I am happy to say that this long standing planned feature has been implemented.

Below are two examples that would have been out of reach before this release:

The left example is a 1024x1024 RGBA image with a slider to scrub the value of its green channel interactively. The right example is a 512x512 scalar image (colormapped in browser) that is blurred using Numba code in response to the slider. In case you were wondering, a fast, interactive image processing web app is about twenty five lines of pure python:

image = scipy.misc.ascent().astype(np.int32)[::-1, :]
w, h = image.shape

source = ColumnDataSource(data=dict(image=[image]))

p = figure(x_range=(0, w), y_range=(0, h))
p.image('image', x=0, y=0, dw=w, dh=h, palette=gray(256), source=source)

def blur(outimage, image, amt):
    for i in range(amt, w-amt):
        for j in range(amt, h-amt):
            px = 0.
            for iw in range(-amt//2, amt//2):
                for jh in range(-amt//2, amt//2):
                    px += image[i+iw, j+jh]
            outimage[i, j] = px/(amt*amt)

def update(attr, old, new):
    out = image.copy()
    blur(out, image, 2*new + 1)[out])

slider = Slider(title="Blur Factor", start=0, end=10, value=0)
slider.on_change('value', update)

curdoc().add_root(column(p, slider))

That’s it!

Some Release Notes

In order to make some improvements for the long term, a few small, immediate breaking changes were unavoidable. It is not expected they will affect many users, but please consult the Release Notes carefully.

As of this Release, Python 3.4 support is no longer maintained under test. Python 2.7 and 3.5+ are the officially supported versions. It is very likely that Bokeh will continue to function as expected with Python 3.4 but we no longer run the continuous testing suite against it.

Lastly, a very loud warning has been added to session.loop_until_closed. It was never the intention that using bokeh.client to run application code outside a Bokeh server be considered a “real” way to do anything. However, mentioning this on the mailing list and Stack Overflow has not been sufficiently discouraging. The new warning outlines the many drawbacks of using bokeh.client in this way. To drive the point home all models/server and plotting/server examples will be removed from GitHub after this release.

Next Steps

The only “must-do” task left for 1.0 is to port BokehJS to TypeScript. Once that is complete, we will immediately plan 1.0 to be the next release. There are some additional “nice-to-have” tasks that may go into 1.0, if time and resources permit:

  • Scriptable animations and transitions
  • New events for document init, busy/done, etc.
  • Support “Patches with holes” for GIS usage
  • Publish BokehJS reference documentation

In order to avoid the overhead and risk of “big releases”, we plan to have more, smaller releases until 1.0 is ready.

As we get closer to a Bokeh 1.0 release, I’d like to thank the 250 total contributors who have helped make Bokeh such an amazing project. As always, for technical questions or assistance, please post to the Bokeh Mailing List and the General Chat Channel. Or, if you’re interested in helping out, drop by the Dev Chat Channel!


Bryan Van de Ven