#2175 & #3723 - Multicategory Multilevel 2+ & Sorting Multicategory #6327
Conversation
|
@richardnm-2 thanks so much for opening this PR! Just a couple of structural comments while I'm reviewing the rest of the code:
|
| return newMatrix; | ||
| } | ||
|
|
||
| exports.transpose = transpose; |
There was a problem hiding this comment.
I haven't fully worked through what you're doing with this, but note there's also a matrix module with its own transposition function, transposeRagged. If that routine doesn't do what you need, let's move this transpose into the matrix module and call it perhaps transposeSquare. Same goes for other functions here that aren't specific to traces per se.
| if(d) return d; | ||
| if(axLabels.length) { | ||
| ax.levelNr = axLabels[0].length; | ||
| ax.levels = axLabels[0].map(function(_, idx) {return idx;}); |
There was a problem hiding this comment.
ax is part of gd._fullLayout and it's therefore a kind of "cleaned" and "completed" version of gd.layout, the layout portion of the figure. With that in mind, it's OK to add things to ax that don't exist in layout.xaxis etc, but only if you prefix them with an underscore.
| var axLabels = []; | ||
| var fullObjectList = []; | ||
| var cols = []; | ||
| // Don't think that the trace should be drawn at all if the lengths don't match. Removing the arrays length check. It is better to fail loudly than silently. |
There was a problem hiding this comment.
For better or worse that's not the approach plotly.js takes. Partly because it gets used in manual data entry applications and other cases where a plot gets drawn before the data is complete, we almost always do our best to NOT fail, and just interpret as much of the data as we can and draw that.
|
|
||
| if(trace.type === 'ohlc' | trace.type === 'candlestick') { | ||
| var sortedValsTransform = sortLib.transpose(axVals); | ||
| gd._fullData[i].open = sortedValsTransform[0]; |
There was a problem hiding this comment.
Similar to my previous comment about adding things to ax, we can't be overwriting contents of gd._fullData in a way that conflicts with the contents of gd.data, but if necessary we can add underscore-prefixed attributes.
Also: this section looks a bit too tightly coupled with concerns that belong to individual trace types. If we need to have logic here that's specific to one or two trace types, that functionality should go in the trace module and get called from there, something like:
var traceSetup = gd._fullData[i]._module.setupMultiCategory;
if(traceSetup) {
traceSetup(gd._fullData[i], sortedValsTransform);
}Then candlestick and ohlc would opt into this behavior by defining a setupMultiCategory function in their index files
|
@richardnm-2 Hello, great to see that someone is working on this and tries to provide a solution :) It is quite a pain with multicategory axis right now. Sorting breaks whenever an item is missing in one stack for example... Hope there will be a fix soon. fyi: We are using plotly in R |
|
@richardnm-2 It's fantastic to see your efforts to improve the current handling of multicategory axes. Currently, when a category lacks data points, it can significantly disrupt the visualization integrity. Also, having more than 2 levels is a must for most of modern data analysis. Is this feature still being worked on at the moment? |
|
Is this close to merging? Would be a great feature to have. |
|
@richardnm-2 May I inquire about the approach you used to resolve this issue in plotly.py? |
|
Hi, is this PR still in progress? Looking forward for this functionality to be introduced in the library. |
|
I really hope this feature will be ok |
#2175 requests for a multicategory multilevel (2+). This is great for data visualization on, as mentioned, "Variability Charts".
In order to expand the capabilities of multicategory axis type, the list structure previously implemented has been changed at
set_convert.jsax.setupMultiCategory(), adding a new file tosrc/lib->sort_traces.js. Working with the traces as a list of objects instead, to try to provide the same data structure tosetCategoryIndex()seemed easier. Sorting is performed reversed, in order to achieve a stable sorting. I don't know about the implementation, but the result is the same as if sorted backwards in pandas (Python), or an Excel table.The capabilities upgrade was then achieved, and the other changes were to make tha axis draw the new data shape, as pushing multiple functions to
seq, in a loop, to account for every level. Also, category indexes had to be updated to comply with the new data shape.The categorical data on the axis was drawn accordingly, but the sorting issue, #3723 showed that this was a known problem. Testing showed that providing the data already sorted to the newPlot function outputed the desired result. So, as the trace levels were already beein looped inside
ax.setupMultiCategory(), and the object list was easyly sortable, before transforming it to a matrix again, it made sense to also update gd._fullData with the sorted data. The place to do so may not be the best and some thought can be given to that.