support plotting multi-dimensional data

[keewis]

• follow-up to follow-up tasks for explore #77
• Tests added
• User visible changes (including notable bug fixes) are documented in changelog.md

This adds sliders to support the visualization of multi-dimensional data (developed during the remote sprint this morning with @ofk123 and @evenmm).

The remaining things to improve are:

• a text box for fine control of the selected value (the slider is a bit short) looks like this is not necessary, we can go step-by-step using the arrow keys
• label-based selection instead of index-based selection
• a decision on where to put the controls (they're below the map for now)
• change the pixel inspection table to include the newly selected values

Follow-up tasks:

• add an example to the lonboard integration gallery

For an example, try:

import xdggs

ds = xdggs.tutorial.open_dataset("air_temperature", "h3").load().dggs.decode()
ds["air"].dggs.explore(alpha=0.8)

cc @kylebarron

[kylebarron]

Cool! Do you have any screenshots/screencasts?

You could add this/your existing integration to the lonboard docs if you'd like https://developmentseed.org/lonboard/latest/examples/#integrations

[keewis]

You could add this/your existing integration to the lonboard docs if you'd like

I'll do that!

I'd like to make sure this doesn't do anything weird with lonboard's layer object first, though. In particular, I'd like to update the values in the layer table with the currently selected ones (so the tool tip shows the currently selected values). Do you have any tips on how to get that to work? Setting the result of RecordBatch.set_column does not appear to do the right thing.

[kylebarron]

I'd like to make sure this doesn't do anything weird with lonboard's layer object first, though.

I don't think I've ever gotten such updates (where the selection happens on the Python/ipywidgets side, not the JS side) to work cleanly. I tried that in developmentseed/lonboard#610 but the selection updating was pretty flaky.

Setting the result of RecordBatch.set_column does not appear to do the right thing.

Are you setting that onto layer.table? I'm not sure that'll work.

The best option might be to use a DataFilterExtension with boolean values of either 0 or 1.

[keewis]

Are you setting that onto layer.table? I'm not sure that'll work.

Yes, that's what I was talking about (and which definitely does not work, at least currently). The issue with DataFilterExtension (if I understand it correctly) is that the data I'd be interested in plotting may potentially be impossible to load into (browser) memory: this might be TB of climate model data that would be computed / loaded into memory on the fly (using dask, for example).

For the example above the data is a couple of megabytes in total, but so far this has been very smooth. That was on a local jupyterlab, though, so I'll have to experiment with a remote jupyterlab and bigger data.

[keewis]

I think we can go ahead and merge this, even though there's still a few things to resolve (the most important thing being the table updates)