Integrate non-geospatial plots with JupyterGIS

[mfisher87]

Use case

We need a user story for integrating non-geo plotting with JupyterGIS. Right now JGIS is entirely map-based, but there's a need to integrate non-map plots to users to explore deeper than they can with a map only. For example, if they want to scatter plot two variables to examine correlation between those variables. Ideally, we also will have glue-like functionality to enable users to associate data on a scatter plot with data on a map. And ideally, we would enable many more kinds of visualizations than scatter plots.

Questions:

• How do we do this in a generic way instead of needing to manually grow the set of visualizations we support over time?
• How do we enable users to decouple their visualizations from JupyterGIS (e.g. a cartokit-like direct manipulation/code generation workflow?)

Example (we probably shouldn't try to do all this all in one PR, this is an example of a complete workflow):

Visualize plots in the JupyterGIS window. Provide new UI elements to store the plots the user has created and allow them to select a plot to display it. Allow the user to drag the plot around so it's reasonably positioned for their needs. Maybe allow multiple plots on the screen at a time. When the user hovers over e.g. points in a scatter plot, the same features on the map are highlighted. Allow the user to right-click a point and zoom to it on the map. Allow the user to change the plot type to e.g. a box plot, line graph, etc. Allow graphic data from multiple layers on the same plot. Allow the user to right-click the plot and "export as Python" to generate an identical plot in a Notebook.

Preferred solution

No response

Alternative solutions

No response

[nakul-py]

Hey @mfisher87 I think we can done like this

In our first PR we can.

• Introduce a new visualization/ module.
• Add support for a single visualization type: Scatter Plot.
• Add a “Create Scatter Plot” action in the layer context menu.
• Implement a simple dialog to select X and Y fields.
• Store a minimal visualization spec in the shared model.
• Render the plot as a draggable overlay inside the main map view.

---

For future PRs

• Linked map ↔ plot selection/highlighting.
• Multiple visualization types.
• Multi-layer plotting: Box plot, Line plot.
• Export to Python.
• Advanced layout management (docking, resizing, toggling).

What you think. If you want we can also shorten our first pr.

[mfisher87]

I think all that sounds great, except I think we need an option to close and re-open the plot in the first PR or users who build plots could be frustrated that it's in the way. Maybe a new tab on the right panel with a list of plots. And plots can have names for easy reference in the right tab!

I think the second PR on this feature should be export to Python, that's a user experience that really needs exploration, and the value of exploring that user experience at this stage is greater than those other features IMO!

[martinRenou]

we also will have glue-like functionality

https://github.com/QuantStack/glue-jupyterlab 😏

I'd love for us to not reinvent the wheel on this. We should adopt a widely used schema-based plotting library for this. Something like vega-lite or similar.

I believe we should also continue with the "collaborative" promise.