point-cluster for scattergl #2493
Conversation
| // create scatter instance by cloning scatter2d | ||
| scene.select2d = createScatter(fullLayout._glcanvas.data()[1].regl, {clone: scene.scatter2d}); | ||
| // create select scatter instance by cloning scatter2d | ||
| scene.select2d = createScatter(fullLayout._glcanvas.data()[1].regl); |
There was a problem hiding this comment.
Since we reuse clustering, creating new scatter is way faster, we don't need to clone data.
There was a problem hiding this comment.
... which makes switching from zoom/pan to select/lasso drag modes waaay faster. Thanks!
|
Thanks @dfcreative ! We should probably apply those commits on top of #2492 to avoid conflicts. |
src/traces/scattergl/index.js
Outdated
| if(errorOpts.copy_ystyle) { | ||
| errorOpts = trace.error_y; | ||
| } | ||
| if (!trace.unselected) unselectedOptions.opacity = DESELECTDIM; |
There was a problem hiding this comment.
Why do we need this all of a sudden?
|
I'm seeing ~200ms improvements in first-render speed at 1e6 pts (from 1400-1500 down to ~1200ms) which is pretty amazing 🎉 That said, on my machine drag and selection appear slower off this branch. Maybe one or a few dependencies haven't been updated properly. Oh wait! Yeah, this branch is still using the |
|
with d972023, the first-render improvements are even better, I'm getting <1000ms first-render time at 1e6 pts which is nothing short of outstanding 🎉 That said, yeah I'm still experiencing slow drag and selections. @dfcreative maybe we'll need to update parts of update routine. |
|
@dfcreative I made scattergl-dry-convert...point-cluster-dev which I think captures of the required updates in scattergl and scatterpolargl of this branch and applies them to #2492 to avoid conflicts down the road. I suggest closing this PR and working off |
|
@dfcreative Here are some problems I noticed off scattergl-dry-convert...point-cluster-dev :
|
|
as for the jasmine tests, a few tests (6 in total) are failing due to the |
This PR includes improvement on regl-scatter2d & introduces point-clustering.
Btw selection, panning & updating are significantly faster.
Clustering is somewhat faster.
Introducing async clustering - later.
cc @etpinard