I found that the bottleneck was computation of the overlapping rectangles every time downsampling scaled changed. When I removed that, and just did simple rendering of each datapoint separately the performance was a lot better - mostly eliminated all the delays when changing downsampling scales.
I also changed the offscreen canvas size from 20000 to 2000, and that improved performance.
Finally, I modified the downsampling strategy to use the MAX rather than the AVG over each slice, and this allows to see more detail when zoomed out.
I deployed the modified version as gs://figurl/franklab-views-dev1g (rather than 1f)
@jsoules
I found that the bottleneck was computation of the overlapping rectangles every time downsampling scaled changed. When I removed that, and just did simple rendering of each datapoint separately the performance was a lot better - mostly eliminated all the delays when changing downsampling scales.
I also changed the offscreen canvas size from 20000 to 2000, and that improved performance.
Finally, I modified the downsampling strategy to use the MAX rather than the AVG over each slice, and this allows to see more detail when zoomed out.
I deployed the modified version as
gs://figurl/franklab-views-dev1g(rather than 1f)https://www.figurl.org/f?v=gs://figurl/franklab-views-dev1g&d=sha1://aa4ce75d420bdedb3e74a3f70378f7747bafa14e&label=03f1-causal-posterior&s=%7B%7D