Use dynspread=True by default for explorer if datashade/rasterize

[ahuang11]

I think it's more user friendly if dynspread is True, or else can barely see anything:

With dynspread=True:

[maximlt]

Pinging @jbednar for feedback!

Seems sensible in the shared example, except that this example has very little data so rasterizing it isn't very relevant.

[ahuang11]

"Seems sensible in the shared example, except that this example has very little data so rasterizing it isn't very relevant."

Right. I was imagining a scenario where a user, who doesn't have much experience in the HoloViz ecosystem, randomly clicking things and seeing what it does. If they see a blank plot, they might be turned away from using datashader, but if they at least see something, that might pique their curiosity.

I suppose something else that can remedy this is having doc specified for each param and so it shows up as a tooltip.

[maximlt]

I think dynspread has docs, except that Bokeh doesn't support tooltips for all the widgets?

[maximlt]

Alternatively, the explorer could warn the user when they set rasterize/datashader on a dataset that has too little data.

[ahuang11]

Is there a downside to setting dynspread=True in any case though?

[maximlt]

Maybe, if so that would be the default in hvPlot?

[ianthomas23]

You wouldn't want to use any form of spreading if you are rendering antialiased lines in datashader. But I don't know if antialiased lines have reached hvplot yet.

[jbednar]

I think dynspread should default to True in hvPlot in general, not just for Explorer, when doing rasterize on a Points or Scatter plot, unless people have passed in an explicit point size. It's not appropriate for a line plot, where instead we should be respecting the line_width option already.

[jbednar]

BTW, I think Explorer should also be enabling either LTTB or rasterize automatically when the data size is too large. Maybe LTTB by default for line data >100000 points, and rasterize by default for point data > 100000 points?

[ahuang11]

What's LTTB?

Oh Largest-Triangle Three-Buckets algorithm