On-par autorange for scattergl #2404

etpinard · 2018-02-26T17:51:36Z

fixes #2354 - to be merged into #2388

This gets scattergl autorange to parity with scatter by reusing its calc-step logic straight-up for data arrays of length less than 1e5. I made a few more improvements to Scattergl.calc that will be highlighted in 0aa0f5e.

I also made a linting commit 5316c47 mostly renaming container -> gd making scattergl/index.js look a little more like the rest of our codebase. I hope @dfcreative won't mind.

... replot is sufficient since regl push.

... to make things look a little more like the rest of plotly.js

- reuse scatter axis-expansion logic - improve 'fast' axis expand routine (using average marker.size as pad value) - use ax.makeCalcdata for all axis types (this creates a new array for linear axes, but makes thing more robust) - add a few TODOs

- most notable change is in gl2d_axes_label2 that now shows the correct to-zero autorange.

etpinard · 2018-02-26T17:55:00Z

src/traces/scattergl/index.js

+ // regl-scatter2d uses NaNs for bad/missing values
+ //
+ // TODO should this be a Float32Array ??
+ var positions = new Array(count2);


@dfcreative could we set up positions with a Float32Array here?

Float32 is not enough precision for some timedate and precise linear/etc ranges.

etpinard · 2018-02-26T17:59:47Z

src/traces/scattergl/index.js


- var x = xaxis.type === 'linear' ? trace.x : xaxis.makeCalcdata(trace, 'x');
- var y = yaxis.type === 'linear' ? trace.y : yaxis.makeCalcdata(trace, 'y');


IMPORTANT with typed array support in mind (-> #2388), I made all traces pass through makeCalcdata unlike previously where x/y array on linear axes bypassed it. Please note that makeCalcdata creates a new array (i.e. x isn't the same as trace.x), unless trace.x is a typed array. So memory-conscious user should switch to using typed arrays.

Note that at 1e6 on my setup, one makeCalcdata call clocks in at roughly 25ms. So using typed arrays can save about 50ms

Even with the extra arrays, memory consumption appears fine. I wouldn't having @dfcreative double-check though.

I guess there are situations we accept now that would have broken previously (but only on linear axes) - the "junk" characters stripped by cleanNumber - worth a test case?

Yes. Good call!

Added mock in c15722f.

On master, the new mock would have looked like:

etpinard · 2018-02-26T18:14:15Z

For reference, here's a few benchmarks I took down while working on this PR:

At 1e6 points:

xy makeCalcdata: ~ 50ms (this can be 🔪 by using typed arrays)
generating the kdtree: ~ 200-300ms (which @dfcreative is hoping to 🔪 soon)
autorange (the fast/approximative routine): < 5ms
total first render time: 1200-1500ms (with set ranges or not)

At 1e5-1 points:

xy makeCalcdata: < 5ms
generating the kdtree: ~20ms
autorange (the slow but on-par with svg routine): < 100-120ms (can be 🔪 when setting axis ranges)
total first render time: 300ms

A few more notable benchmarks:

At 1e4 points, first render time for scattergl is ~150ms (vs ~800ms for svg scatter).
the slow but on-par with svg autorange routine would clock in at ~800ms at 1e6 points. So yeah, using a fast approximative version for big data array is still the right call.

I used Chrome 64 on a Lenovo t450s with Ubuntu 16.04.

alexcjohnson · 2018-02-27T14:50:16Z

src/traces/scattergl/index.js

- x[i] = i;
+ if(xa.type === 'log') {
+ for(i = 0; i < count2; i += 2) {
+ positions[i] = xa.d2l(positions[i]);


🐎 we've already been through makeCalcData so we should be able to do c2l here

Thanks for pointing this out! Done in -> 98d2407

alexcjohnson · 2018-02-27T15:20:19Z

src/traces/scattergl/index.js

+ calcAxisExpansion(gd, trace, xa, ya, x, y, ppad);
+ } else {
+ if(markerOptions) {
+ ppad = 2 * (markerOptions.sizeAvg || Math.max(markerOptions.size, 3));


How did you find sizeAvg to work in practice? A little hard to say I guess, since we don't have a lot of real data > 1e5 points to play with... You're right that we don't want to just use sizeMax, it's worth accepting a bit of clipping in order to generally have less wasted space, and much of the time the largest point won't be on any edge... just wondering how that balance plays out in practice, whether we would be better off with something like halfway between the average and max.

Anyway perhaps we don't have a god way to answer that question right now. I'll just mention that in case we do want to try and do better later, we could find some heuristics that only add a little bit of computation, like binning data points into top/middle/bottom thirds, and only using the top third for the top padding... maybe even with a smooth weighting of the size based on how far it is from the edge. That would still be far faster than the full calculation but could do a better job reducing wasted space without too much clipping.

alexcjohnson · 2018-02-27T15:23:33Z

test/jasmine/tests/gl2d_plot_interact_test.js

+ })
+ .then(function() {
+ expect(gd._fullLayout.xaxis.range).toBeCloseToArray(glRangeX, 'x range');
+ expect(gd._fullLayout.yaxis.range).toBeCloseToArray(glRangeY, 'y range');


Beautiful test!

dy · 2018-02-27T18:33:06Z

src/traces/scattergl/index.js


 // prepare colors
- if(multiMarker || Array.isArray(markerOpts.color) || Array.isArray(markerOpts.line.color) || Array.isArray(markerOpts.line) || Array.isArray(markerOpts.opacity)) {
+ if(multiSymbol || multiColor || multiLineColor || multiOpacity) {


dy · 2018-02-27T18:36:48Z

test/jasmine/tests/gl2d_plot_interact_test.js

+ {name: 'annotations', fig: require('@mocks/gl2d_annotations.json')}
+ ];
+
+ specs.forEach(function(s) {


etpinard · 2018-02-28T00:53:15Z

@dfcreative @alexcjohnson all comments have been addressed.

dy · 2018-02-28T01:01:36Z

LGTM! 💃

alexcjohnson · 2018-02-28T01:34:28Z

💃 from me too! 🎉

etpinard added 6 commits February 26, 2018 11:21

make scattergl dragmode relayout replot not recalc

909120e

... replot is sufficient since regl push.

large lint commit in scattergl/index.js

5316c47

... to make things look a little more like the rest of plotly.js

improvements to Scattergl.calc

0aa0f5e

- reuse scatter axis-expansion logic - improve 'fast' axis expand routine (using average marker.size as pad value) - use ax.makeCalcdata for all axis types (this creates a new array for linear axes, but makes thing more robust) - add a few TODOs

update baselines

39ef5a9

- most notable change is in gl2d_axes_label2 that now shows the correct to-zero autorange.

update jasmine tests for new scattergl auto-ranges

36b4e25

add svg vs gl scatter autorange 🔒s

40f93f7

etpinard added bug something broken feature something new status: reviewable labels Feb 26, 2018

etpinard commented Feb 26, 2018

View reviewed changes

etpinard added this to the v1.35.0 milestone Feb 26, 2018

alexcjohnson reviewed Feb 27, 2018

View reviewed changes

dy reviewed Feb 27, 2018

View reviewed changes

etpinard mentioned this pull request Feb 28, 2018

gl2d known limitations #130

Closed

21 tasks

etpinard added 2 commits February 27, 2018 19:39

use c2l instead of d2l

98d2407

🔪 TODOs

01a8443

etpinard mentioned this pull request Feb 28, 2018

Look for ways to improve big-data axis-expand algo #2417

Closed

add mock 🔒ing down cleanNumber for scattergl

c15722f

etpinard merged commit a2fb88b into typed-arrays-support Feb 28, 2018

etpinard deleted the scattergl-autorange branch February 28, 2018 18:46

This was referenced Feb 28, 2018

scattergl fill - autorange and restyle #2354

Closed

Autosize / autorange fixes #2437

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

On-par autorange for scattergl #2404

On-par autorange for scattergl #2404

etpinard commented Feb 26, 2018

etpinard Feb 26, 2018

dy Feb 27, 2018

etpinard Feb 26, 2018 •

edited

Loading

etpinard Feb 26, 2018

alexcjohnson Feb 27, 2018

etpinard Feb 27, 2018

etpinard Feb 28, 2018

etpinard commented Feb 26, 2018

alexcjohnson Feb 27, 2018

etpinard Feb 28, 2018 •

edited

Loading

alexcjohnson Feb 27, 2018

etpinard Feb 28, 2018

alexcjohnson Feb 27, 2018

dy Feb 27, 2018

dy Feb 27, 2018

etpinard commented Feb 28, 2018

dy commented Feb 28, 2018

alexcjohnson commented Feb 28, 2018


		var x = xaxis.type === 'linear' ? trace.x : xaxis.makeCalcdata(trace, 'x');
		var y = yaxis.type === 'linear' ? trace.y : yaxis.makeCalcdata(trace, 'y');

On-par autorange for scattergl #2404

On-par autorange for scattergl #2404

Conversation

etpinard commented Feb 26, 2018

Choose a reason for hiding this comment

Choose a reason for hiding this comment

etpinard Feb 26, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

etpinard commented Feb 26, 2018

Choose a reason for hiding this comment

etpinard Feb 28, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

etpinard commented Feb 28, 2018

dy commented Feb 28, 2018

alexcjohnson commented Feb 28, 2018

etpinard Feb 26, 2018 •

edited

Loading

etpinard Feb 28, 2018 •

edited

Loading