Bokeh Visualization/Interactive Model Architecture for Multivariate Analysis

Background: I am working on an analysis problem for multivariate modeling and visualization of datasets. Depending on the particular data, a user might be interested in relationships between two variables; in other cases, it could be five, for example. All that is to say it will be application and problem-dependent.

Question: My understanding is that, historically, the guidance for bokeh visualizations is to generate the models (figures, widgets, etc.) upfront based on the largest problem size one expects to solve and then disable or hide those models not needed in a particular scenario rather than dynamically adding and removing figures, widgets, etc. as the problem size changes.

Is this still the recommended approach? Is there anything in the bokeh roadmap geared towards cleanly and efficiently adding and removing models from a document dynamically?

That sounds like a good use case for Panel, especially when combined with HoloViews or hvPlot.

1 Like



For all of my other applications to date, I have been working exclusively with bokeh.

I was under the impression, perhaps incorrectly, that HoloViews is a wrapper to bokeh that provides a lot of convenience capabilities and developer-friendly formatting and layout, but it does not extend the feature set.

I’ll look into the documentation and experiment with it so I can be more informed about what is and what is not possible.

And the Panel solution you mention looks like a nice framework; I’ve seen mention of it but hadn’t actually looked for it until you provided the link.

Thanks again.


Just making my initial detailed look into Panel. Thanks for the suggestion; it looks like it will support the dynamic layout changes I need, and its bokeh server foundation is another plus for my application. Looks like a very promising way forward.

Thanks again.