Add `density_scatter_plotly`

[janosh]

one very important plot type that i've been meaning to port over to pymatviz is a truly scalable version of density_scatter but using plotly as backend to get interactive tooltips. those can be very useful when making parity plots of e.g. of machine learning models tested on large datasets:

density_scatter could either be upgraded by adding a backend="plotly" option or by making a new plot function density_scatter_plotly. i prefer the 1st option from a user-facing API perspective but think the code would end up more readable taking the 2nd option. still, option 1 probably better.

given matbench discovery already has several examples of how to make such plots, it would largely be a case of wrapping the way bin_df_cols is used here into a more streamlined user-friendly function.

[janosh]

@DanielYang59 let me know if this is something you'd be interested to collaborate on?

[DanielYang59]

Certainly. But shall we finish #157 before this (one thing at a time)? I'm currently taking some time off, and would get my hands on #157 very soon :)

density_scatter could either be upgraded by adding a backend="plotly" option or by making a new plot function density_scatter_plotly. i prefer the 1st option from a user-facing API perspective but think the code would end up more readable taking the 2nd option. still, option 1 probably better.

I don't have extensive experience with plotly, if it shares most grammar with matplotlib, 1st option might be preferred. If not, I would personally prefer 2nd (the ptable_heatmap_plotly for example). Also the 2nd option might be useful if we were to separate pymatviz into matplotlib and plotly at some point (i.e. static plotters and interactive plotters)?

[janosh]

But shall we finish #157 before this (one thing at a time)?

i think those PR shouldn't cause merge conflicts since they touch different files. unless you're worried about context switching? that's a valid point.

I don't have extensive experience with plotly, if it shares most grammar with matplotlib, 1st option might be preferred.

plotly and matplotlib syntax is entirely different. i actually prefer plotly's syntax. it's more consistent across figure types and therefore easier to learn. NOMAD makes extensive use of plotly as well.

Also the 2nd option might be useful if we were to separate pymatviz into matplotlib and plotly at some point (i.e. static plotters and interactive plotters)?

i wasn't planning on that but of course i'm open to arguments

[DanielYang59]

But shall we finish #157 before this (one thing at a time)?

i think those PR shouldn't cause merge conflicts since they touch different files. unless you're worried about context switching? that's a valid point.

Yep I want to focus on one thing at a time. Because I could see quite some work coming for the heatmap plotter refactor :)

I don't have extensive experience with plotly, if it shares most grammar with matplotlib, 1st option might be preferred.

plotly and matplotlib syntax is entirely different. i actually prefer plotly's syntax. it's more consistent across figure types and therefore easier to learn. NOMAD makes extensive use of plotly as well.

Thanks for letting me know. I really want to learn something about plotly from this collaboration :)

Also the 2nd option might be useful if we were to separate pymatviz into matplotlib and plotly at some point (i.e. static plotters and interactive plotters)?

i wasn't planning on that but of course i'm open to arguments

I wasn't suggesting this even for the near future, I would assume it's something we might consider when pymatviz gets really large :)