Bokeh ColumnDataSource from user input

[ahartikainen]

To enable linked brushing, we need an option for a user to input ColumnDataSource object which contains the needed data. This will also enable to combine different plot types.

We need to be explicit what different kinds of ColumnDataSources we need. At least we have

• mcmc sample
• density (from mcmc sample)
• observed (no chain, draw)

Ideally we would like to have some default fields for good hover tooltip. (Like information for draw, chain, variable name(?); dims + coords)

https://docs.bokeh.org/en/1.4.0/docs/reference/models/sources.html

https://docs.bokeh.org/en/1.4.0/docs/user_guide/data.html

[amukh18]

I would like to work on this issue

[amukh18]

@ahartikainen Could you please give me a very brief overview of how the implementation of user input should work?

[amukh18]

So the output should be an InferenceData object which we pass to the plotting functions ?

[ahartikainen]

I will need to show you one example. I think I have that function somewhere (inferencedata to cds)

[amukh18]

This one ?:https://github.com/arviz-devs/arviz/pull/981 I am currently going through this

[ahartikainen]

Yes, that one.

I think we could see if that could be used (or is it already used) in the bokeh plotting function.

If it is already used, we could add a possibility include cds in some dictionary/argument (probably in backend_kwargs) and if present, use it instead of CDS created in the function.

Let's think how to handle density-based plots later.

[amukh18]

I am a little confused. I was looking at the bokeh backend file for plot_trace and I think there are already some lines of code that convert the InferenceData object to ColumnDataSource in that file itself.

[ahartikainen]

Yeah, this is probably much more involved thing that it seems at first, I think I will maybe implement this.

[amukh18]

@ahartikainen Okay

[amukh18]

@ahartikainen Or Is there any way I could implement at least part of this? I think it would help me learn about Arviz's backend system better

[ahartikainen]

Yes, we can implement this together. Let me give you instructions where we start in a few hours

[amukh18]

Thank you! I look forward to working on this issue

[ahartikainen]

Ok, I did some inspection for our plotting code, and it is certainly not ready for this feature.

I think we really need to do some serious rewriting to enable this.

I think it is better to focus on other issues for now.

cc @OriolAbril

[amukh18]

I understand. I will continue my work on the other issues.