Introduce cartesian axis breaks #4614

etpinard · 2020-03-03T20:00:48Z

which presents a solution for removing weekends, holidays and non-work hours as requested in #1382 and lays the groundwork from "visually skipped" ranges as in the screenshot in #4210.

In brief, this PR adds a breaks array attribute to cartesian axes. Each item inside breaks defines an axis break either using a 2-item bounds array or a free-length array of values. When using bounds on date axes, one can set a pattern to generate breaks e.g. over all weekends or non-work hours. When using values, one can set the "spread" of each value using dvalue. The inclusive/exclusive behaviour at the bounds is determined by the operation attributes (as used in filter transforms).

Some examples:

breaks on linear x and y axes: https://codepen.io/etpinard/pen/eYNErKK?editors=1010
breaks using pattern on date axis: https://codepen.io/etpinard/pen/MWwvGBV?editors=1010
breaks using values on date axis: https://codepen.io/etpinard/pen/PoqKedG

To implement axis breaks,

we first set coordinates that fall within the breaks to BADNUM (more info in baf753a). Please note that this implies that lines remain continuous across a break if there's no (x,y) pt inside the break in question.
sanitize the breaks present inside the axis range during ax.setScale (i.e. when the range is known) and adapt the l2p and p2l mappings (more info in 259eafa)
Use the sum of the piecewise data-space ranges between the breaks (instead of simply range[1] - range[0]) for autorange padding, ticks placement computations and during drag interactions.

cc @archmoj @alexcjohnson @nicolaskruchten @antoinerg

... including potential future attributes `gap` and `gapmode`

- used during makeCalcdata to mask (i.e. set to BADNUM) coordinates inside the axis breaks.

- add abstractions: + ax._breaks, the disjoint breaks inside the range though ax.locateBreaks + ax._lBreaks, length of these breaks in value space + ax._m2, l2p slope (same for all intervals) + ax._B, set of l2p offsets (one for each of the (N+1) piecewise intervals) - adapt l2p and p2l for axis breaks - add jasmine tests for setConvert and hover labels

... as the value-space length is reduced when axis breaks are present within the trial range

- reduce the value-space range by `ax._lBreaks` to get the correct the number of auto ticks on the axes. - add logic to remove ticks falling inside breaks - add (could do better) logic for ticks that overlap on either side of a break.

- test breaks on x axes - test breaks on y axes - test date axes with patterns - test breaks with tickvals and ticktext - test for zeroline inside breaks

archmoj · 2020-03-04T15:47:08Z

src/plots/cartesian/axis_defaults.js

+                    itemOut.enabled = false;
+                    return;
+                }
+            }


What about handling autorange: 'reversed' case?
In the following demos I tried to flip the axes which it didn't work.
https://codepen.io/MojtabaSamimi/pen/yLNzLra?editors=0010
https://codepen.io/MojtabaSamimi/pen/bGdoGyG?editors=0010
https://codepen.io/MojtabaSamimi/pen/QWbqWRj?editors=0010

Ah right, I forgot to test those. They won't be handled here as this logic has to do with set (i.e. autorange: false) ranges, but yeah I'll get this fixed. Thanks!

src/plots/cartesian/set_convert.js

archmoj · 2020-03-04T16:45:23Z

src/plots/cartesian/set_convert.js

+                                break;
+                        }
+
+                        while(t <= r1) {


Could we add a finite test for t before entring this loop?
Is there any scenario (e.g. on reversed axes) in which step could be negative?

nicolaskruchten · 2020-03-04T21:07:57Z

Looks great! Do we have any control over the visual aspect of the break? Can we put in a gap or a squiggle or something?

etpinard · 2020-03-04T21:14:42Z

Do we have any control over the visual aspect of the break? Can we put in a gap or a squiggle or something?

Not at the moment.

plotly.js/src/plots/cartesian/layout_attributes.js

Lines 346 to 375 in 187c93a

    
                   /* 
        
                   gap: { 
        
                       valType: 'number', 
        
                       min: 0, 
        
                       dflt: 0, // for *date* axes, maybe something else for *linear* 
        
                       editType: 'calc', 
        
                       role: 'info', 
        
                       description: [ 
        
                           'Sets the gap distance between the start and the end of this break.', 
        
                           'Use with `gapmode` to set the unit of measurement.' 
        
                       ].join(' ') 
        
                   }, 
        
                   gapmode: { 
        
                       valType: 'enumerated', 
        
                       values: ['pixels', 'fraction'], 
        
                       dflt: 'pixels', 
        
                       editType: 'calc', 
        
                       role: 'info', 
        
                       description: [ 
        
                           'Determines if the `gap` value corresponds to a pixel length', 
        
                           'or a fraction of the plot area.' 
        
                       ].join(' ') 
        
                   }, 
        
                   */ 
        
                   // To complete https://github.com/plotly/plotly.js/issues/4210 
        
                   // we additionally need `gap` and make this work on *linear*, and 
        
                   // possibly all other cartesian axis types. We possibly would also need 
        
                   // some style attributes controlling the zig-zag on the corresponding 
        
                   // axis.

nicolaskruchten · 2020-03-04T21:37:39Z

OK. I feel like introducing breaks without visible gaps is a bit scary in other contexts than "drop weekends for business metrics" ... what's the LOE on visible gaps?

etpinard · 2020-03-04T22:02:49Z

what's the LOE on visible gaps?

Probably 1-2 days for gaps w/o squiggle on the axis. I suspect that getting everyone happy with the squiggle look/behaviour will take a few days on its own.

src/plots/cartesian/layout_attributes.js

nicolaskruchten · 2020-03-05T13:57:15Z

Zooming way in to the pattern date mock yields some odd tick labels:

src/plots/cartesian/layout_attributes.js

- pass axis breaks to mock rangeslider figures - consider axis breaks in rangeslider d2p & p2d - add mocks + drag test

... that way, positions that include some sort of offset (e.g. bar x0/x1, ohlc xo/xc) have the adequate px position.

nicolaskruchten · 2020-03-05T18:41:15Z

To capture what was discussed today: for 1.53 let's activate this feature for date axes only please. We'll need to think about axis gap and bar charts etc before turning this on on linear/log axes I think.

etpinard · 2020-03-05T21:12:33Z

TODO:

coerce breaks only on type: 'date' axes, adapt tests and mocks!
make sure axis breaks work on axes with range[0] > range[1]

see comments over on: axis-breaks...axis-breaks-fixups

@archmoj will attempt to fix these tomorrow and on Monday the 9th.

better "first tick" algorithm e.g. what to do when the canonical first auto tick falls into a break? See b5aaf92
better "overlapping tick in either side of an axis break fix" algorithm, see comment in 6bec94d

I should be able to look at these when I get back Tuesday the 10th.

- this will be dealt with in #4630

etpinard · 2020-03-12T20:47:17Z

better "first tick" algorithm e.g. what to do when the canonical first auto tick falls into a break?

Commit 28c328d made things a bit better. Still room for improvement though.

Improved once more in 1b55b42

Pan fixes from @archmoj are in 40d57fa

Should we warn the user in that case?
Could we visible: false scattergl on date axes with breaks?

Yes, let's leave gl out of this PR. Hide them - and log a warning - in that case. But @archmoj please make an issue to come back to this soon, I can see this being important for some users.

Done in 49b4053

archmoj · 2020-03-12T20:59:28Z

Great PR!
🎖️ 🎖️ 🥇

Thanks very much @etpinard for tackling all this.
💃 🕺

etpinard · 2020-03-12T21:23:04Z

Ok, merging to get a first round of QA rolling (I suspect a few bug reports will come out of it).

Updated examples:

breaks using pattern on date axis: https://codepen.io/etpinard/pen/PoqRNoL
breaks using values on date axis: https://codepen.io/etpinard/pen/OJVvNJG

nicolaskruchten · 2020-03-13T13:08:40Z

🏆 thanks @etpinard 🙇

kenahoo · 2020-10-01T22:23:23Z

Is this functionality part of any release? Or is it still experimental? I'm having a little trouble finding how to do it in the docs.

archmoj · 2020-10-02T12:21:38Z

Is this functionality part of any release? Or is it still experimental? I'm having a little trouble finding how to do it in the docs.

@kenahoo yes it is available since v1.53.0
You could find javascript documentation at https://plotly.com/javascript/reference/layout/xaxis/#layout-xaxis-rangebreaks.
The documentation for Python at the bottom of this page https://plotly.com/python/time-series/ could also be of interest.

etpinard added 11 commits March 3, 2020 11:49

introduce axis breaks attributes

4f6ea7e

... including potential future attributes `gap` and `gapmode`

add axis breaks default logic

e9cfe04

add ax.maskBreaks method

baf753a

- used during makeCalcdata to mask (i.e. set to BADNUM) coordinates inside the axis breaks.

adapt autorange routine for axis breaks

eca3da7

... as the value-space length is reduced when axis breaks are present within the trial range

adapt calcTicks for axis breaks

6bec94d

- reduce the value-space range by `ax._lBreaks` to get the correct the number of auto ticks on the axes. - add logic to remove ticks falling inside breaks - add (could do better) logic for ticks that overlap on either side of a break.

adapt dragbox logic for axis breaks

fe80cad

do not show zeroline when it falls inside an axis break

d425373

add axis breaks mocks

5f2fbe0

- test breaks on x axes - test breaks on y axes - test date axes with patterns - test breaks with tickvals and ticktext - test for zeroline inside breaks

add TODO for better "first tick" algo on date axes

b5aaf92

fix typo in comment

187c93a

etpinard added type: feature status: reviewable labels Mar 3, 2020

etpinard added this to the v1.53.0 milestone Mar 3, 2020

archmoj reviewed Mar 4, 2020

View reviewed changes

src/plots/cartesian/set_convert.js Show resolved Hide resolved

archmoj reviewed Mar 4, 2020

View reviewed changes

src/plots/cartesian/set_convert.js Show resolved Hide resolved

archmoj reviewed Mar 4, 2020

View reviewed changes

nicolaskruchten reviewed Mar 5, 2020

View reviewed changes

src/plots/cartesian/layout_attributes.js Outdated Show resolved Hide resolved

nicolaskruchten reviewed Mar 5, 2020

View reviewed changes

src/plots/cartesian/layout_attributes.js Outdated Show resolved Hide resolved

etpinard added 3 commits March 5, 2020 09:32

fix axis breaks + rangeslider behavior

ebca01b

- pass axis breaks to mock rangeslider figures - consider axis breaks in rangeslider d2p & p2d - add mocks + drag test

during l2p(v) when v falls into breaks, pick offset closest to it

3957d95

... that way, positions that include some sort of offset (e.g. bar x0/x1, ohlc xo/xc) have the adequate px position.

fix typo in break bounds description

76a265e

etpinard added 2 commits March 12, 2020 16:45

fix 'increase dtick' lgoic for cases with dtick value starting 'M'

1b55b42

set scattergl and splom traces to visible:false on axis with breaks

49b4053

- this will be dealt with in #4630

etpinard mentioned this pull request Mar 12, 2020

support rangebreaks in gl2d traces #4630

Open

etpinard merged commit 7ca1777 into master Mar 12, 2020

etpinard deleted the axis-breaks branch March 12, 2020 21:23

snyk-bot mentioned this pull request Apr 23, 2020

[Snyk] Upgrade plotly.js from 1.52.1 to 1.53.0 mc-ireiser/vue-plotly#1

Open

snyk-bot mentioned this pull request May 1, 2020

[Snyk] Upgrade plotly.js-basic-dist from 1.52.3 to 1.53.0 libamtrack/web#583

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Introduce cartesian axis breaks #4614

Introduce cartesian axis breaks #4614

etpinard commented Mar 3, 2020

archmoj Mar 4, 2020

etpinard Mar 4, 2020 •

edited

archmoj Mar 4, 2020

nicolaskruchten commented Mar 4, 2020

etpinard commented Mar 4, 2020

nicolaskruchten commented Mar 4, 2020

etpinard commented Mar 4, 2020

nicolaskruchten commented Mar 5, 2020

nicolaskruchten commented Mar 5, 2020

etpinard commented Mar 5, 2020 •

edited

etpinard commented Mar 12, 2020

archmoj commented Mar 12, 2020

etpinard commented Mar 12, 2020 •

edited

nicolaskruchten commented Mar 13, 2020

kenahoo commented Oct 1, 2020

archmoj commented Oct 2, 2020

Introduce cartesian axis breaks #4614

Introduce cartesian axis breaks #4614

Conversation

etpinard commented Mar 3, 2020

archmoj Mar 4, 2020

Choose a reason for hiding this comment

etpinard Mar 4, 2020 • edited

Choose a reason for hiding this comment

archmoj Mar 4, 2020

Choose a reason for hiding this comment

nicolaskruchten commented Mar 4, 2020

etpinard commented Mar 4, 2020

nicolaskruchten commented Mar 4, 2020

etpinard commented Mar 4, 2020

nicolaskruchten commented Mar 5, 2020

nicolaskruchten commented Mar 5, 2020

etpinard commented Mar 5, 2020 • edited

etpinard commented Mar 12, 2020

archmoj commented Mar 12, 2020

etpinard commented Mar 12, 2020 • edited

nicolaskruchten commented Mar 13, 2020

kenahoo commented Oct 1, 2020

archmoj commented Oct 2, 2020

etpinard Mar 4, 2020 •

edited

etpinard commented Mar 5, 2020 •

edited

etpinard commented Mar 12, 2020 •

edited