Axis name breaks chart on HTML export but not in notebook

[dcbb]

I have built a chart where renaming an axis breaks the HTML export of the chart, but not the display in a Jupyter notebook. When I open the exported HTML file, the error message from the error handler is displayed. Strangely, when I copy the pure chart dictionary from the HTML into the Vega Lite web editor, things work and the chart displays fine.

Apologies for the complex example. I could not reproduce the error with a simpler one, and I need the data generator because I cannot share my original data.

This is the version, where the output (test.html) is broken, while the chart displays fine in a notebook:

    def generate_data():
        N = 100
        for geo in 'ABC':
            for platform in 'PQ':
                y = np.cumsum(np.random.normal(size=N))
                x = np.arange(N)
                data = pd.DataFrame({'x': x,
                                     'mean': y,
                                     'q25': y - 0.1*x,
                                     'ch': y + 0.1*x,
                                     'cl2': y - 0.05*x,
                                     'ch2': y + 0.05*x,
                                    })
                data['geo'] = geo
                data['platform'] = platform
                yield data
                
    chart = alt.Chart().mark_area(opacity=0.3).encode(x='x:Q', y='q25:Q', y2='ch:Q')
    chart += alt.Chart().mark_area(opacity=0.3).encode(x='x:Q', y='cl2:Q', y2='ch2:Q')
    chart += alt.Chart().mark_line().encode(x='x:Q', 
                                            y=alt.Y('mean:Q', title='RENAMED'), 
                                            tooltip=['geo:N','x:Q','mean:Q'])
    chart = chart.facet(data=pd.concat(list(generate_data())), row='geo:N', column='platform:N')
    chart.save(io.reports_path('test.html'))
    chart

The export works again when I replace alt.Y('mean:Q', title='RENAMED') in the mark_line() part with 'mean:Q', not renaming the axis.

I'm working on fresh (yesterday) Conda install of Altair.

[jakevdp]

It might be that you are using a frontend renderer (e.g. JupyterLab/Jupyter notebook) that has an older version of Vega-Lite, which doesn't support title as an encoding argument.

What happens if you change y=alt.Y('mean:Q', title='RENAMED') to y=alt.Y('mean:Q', axis=alt.Axis(title='RENAMED'))?

[jakevdp]

If you open the javascript console, I suspect you'll see an error along the lines of

 The input spec uses vega-lite v2.5.2, but the current version of Vega-Lite is 2.4.1.

if so, this is the cause of the problem. title as an encoding argument was added in vega-lite 2.5. Note that the version of Vega-Lite used to render Altair plots is not controlled by Altair, but by the renderer (e.g. the notebook extension or jupyterlab extension) so you'll have to update that.

[dcbb]

@jakevdp The change you proposed fixes the export!

Also, in the HTML generated by Altair's save() method, we have:

 <script src="https://cdn.jsdelivr.net/npm//[email protected]"></script>

When I change this manually to

 <script src="https://cdn.jsdelivr.net/npm//[email protected]"></script>

things work with the original code I posted (with the title parameter).

Please note that the chart renders well in the notebook, also with the title parameter – how can this be if I have wrong version of Vega lite?

Also, where can I update my Vega lite version?

I just conda-installed everything, and I'm not really sure which installer installed which version of vega-lite, to be honest. A conda list | grep vega just gives me

 vega                      1.3.0                    py36_0    conda-forge
 vega_datasets             0.5.0                      py_0    conda-forge

[jakevdp]

The change you proposed fixes the export!

Great!

Please note that the chart renders well in the notebook, also with the title parameter – how can this be if I have wrong version of Vega lite?

As I mentioned above, the version of Vega-Lite used by your frontend is not controlled by Altair, but by the frontend renderer.

where can I update my Vega lite version?

If you're using Jupyter Notebook (not JupyterLab), the version of Vega-Lite used to render plots is defined in the vega extension, namely in https://github.com/vega/ipyvega/blob/master/package.json via https://github.com/vega/vega-embed/blob/master/package.json. I wish that were easier to access from Python; I've opened an issue regarding that: vega/ipyvega#97

If you update the vega package from pip, it should use the most recent vega-lite version.

For HTML export, the vega-lite version is defined by Altair itself, namely here: https://github.com/altair-viz/altair/blob/master/altair/vegalite/v2/schema/__init__.py#L4
You can access it with

>>> import altair
>>> altair.VEGALITE_VERSION
'2.5.2'