violin trace: why an area is calculated if all values are 0?

[micosacak]

Please see my question in stackoverflow

I have an array of values from 0-10 with and 20 groups in each. For some groups or categories all values are 0. When I use violin plot, there is an area calculated which is below 0, but I want to get thedifferent that I got from Seurat package VlnPlot. Is there a way to overcome that problem? Here is a part of java script code:

var my_violin = [];
var vln_layout = {};
var my_violin =[
{
    type:"violin",
    x:xvalues,
    y:my_categories,
    points:"all",
    pointpos:0,
    color:"black",
    width:0.75,
    marker:{color:"black",size:4},
    jitter:1,
    span:0,
    transforms:[
    {
        type:"groupby",
        groups:my_categories,
        styles:my_styles
    }
    ]
}
]

var vln_layout = {title: "my title",    xaxis:{showline: false,showgrid: false,zeroline: false,
    categoryorder: 'category ascending'}, yaxis: {showline: false,showgrid: false,zeroline:false},
    showlegend:true,legend:{traceorder:'normal',categoryorder:"ascending"}};

Plotly.newPlot('myDiv2', my_violin, vln_layout, {showSendToCloud: false});

[antoinerg]

Thank you @micosacak for your interest in plotly.js and for reporting this.

When reporting an issue, it is more convenient for the maintainers if you provide a Codepen showing the problem. I prepared one here: https://codepen.io/antoinerg/pen/QoqWBb. Can you fork it and add the missing variables xvalues, my_categories and my_styles?

Thank you!

[micosacak]

Thanks a lot @antoinerg. I forked the values as following;

xvalues and yvalues. my_categories is equal to xvalues. I changed it accordingly.

[micosacak]

And especially, Cell_0, Cell_5, Cell_6, Cell_7 and Cell_17 have zero values for all. I also wonder why Cell_14 and Cell_16, as well with negative values.

[antoinerg]

Trying to plot datasets [0,0,0,0] (left) and [0,0,0,1] (right) with violins:

• using seaborn:
  

• using plotly.js:
  https://codepen.io/antoinerg/pen/oVGxvN
  

Thanks @micosacak for reporting this issue!

[antoinerg]

Even when setting spanmode to hard, the calcSpan function returns [-1, 1] even if all values are zeroes. This happens when calling dummyAx.cleanRange(); on line 155:

plotly.js/src/traces/violin/calc.js

Lines 126 to 158 in 7e68a2e

	function calcSpan(trace, cdi, valAxis, bandwidth) {
	var spanmode = trace.spanmode;
	var spanIn = trace.span || [];
	var spanTight = [cdi.min, cdi.max];
	var spanLoose = [cdi.min - 2 * bandwidth, cdi.max + 2 * bandwidth];
	var spanOut;
	function calcSpanItem(index) {
	var s = spanIn[index];
	var sc = valAxis.type === 'multicategory' ?
	valAxis.r2c(s) :
	valAxis.d2c(s, 0, trace[cdi.valLetter + 'calendar']);
	return sc === BADNUM ? spanLoose[index] : sc;
	}
	if(spanmode === 'soft') {
	spanOut = spanLoose;
	} else if(spanmode === 'hard') {
	spanOut = spanTight;
	} else {
	spanOut = [calcSpanItem(0), calcSpanItem(1)];
	}
	// to reuse the equal-range-item block
	var dummyAx = {
	type: 'linear',
	range: spanOut
	};
	Axes.setConvert(dummyAx);
	dummyAx.cleanRange();
	return spanOut;
	}

I haven't delved into violin much so I am not sure what is the best way to tackle the issue at hand. Any suggestions @plotly/plotly_js ?

[antoinerg]

@alexcjohnson I'm tempted to return a tiny range around cdi.min and cdi.max when they're equal (say +/- 1E-6).

It looks fine alongside another violin:

but when displayed alone, it has an arbitrary size:

[alexcjohnson]

I don't think we can do that, whatever size we choose someone will come up with a data set that has a natural size smaller than that. It has to actually have zero size, even if that means a special case with its own drawing code.

[antoinerg]

@alexcjohnson I don't know what I was thinking 😅! That would have been a horrible hack. PR #3625 is much better.