missing groups

[sleepryansleep]

Hi, I have the following exclusive (disjointed) sets:

A : 7516
A&B : 781
A&B&C : 324
A&B&C&D : 10336
A&B&D : 2525
A&C : 817
A&C&D : 8847
A&D : 6418
B : 7621
B&C : 369
B&C&D : 1149
B&D : 1465
C : 3152
C&D : 4118
D : 26642

Depending on whether I plot as ellipses or circles, there is some number of overlap areas missing. I guess that there is some kind of trade being made by eulerr between approximating areas and if some cutoff is made (ie the areas are too far from representing the true areas) then that overlap is left out. For reference, the webtool at https://www.meta-chart.com is able to plot this data (though likely imperfectly). So my questions are 1. is there a way to adjust this cutoff to be less stringent? ie force it to plot all areas, despite loss in accuracy? 2. is there a better R tool for what I am trying to accomplish? 3. is there some work-around, such as rounding / adjusting the data myself? I need this to be able to fit into a shell where the number of overlaps will differ between experiments. Thank you!

[jolars]

It's commonly the case that areas are missing for combinations with more than 3 sets. eulerr tries to minimize an objective and if leaving out some combinations helps to minimize that objective, then that will typically happen. It looks like meta-chart does venn diagrams, which are not the same as area-proportional euler diagrams.

Have you tried to modify loss and loss_aggregator?

[sleepryansleep]

I've tried loss and loss_aggregator, they don't help too much. Still losing between 2-3 regions. Here is what meta-chart gives me. I don't know exactly what residuals they have, but this doesn't seem awful.

[jolars]

I see. Yeah, it's hard to tell. It looks to me that this could be coming from https://github.com/benfred/venn.js. So maybe https://github.com/montilab/vennr could be what you want? If it turns out that venn.js does produce better results then I'd be happy to know about that!

In general, this feature, forcing intersections to be present, is something that mny users request and it would be nice to cater to that. One possibility would be to precompute layouts for all the possible combinations of intersections and use that as a starting layout and use a loss function that would tend to infinity if any of the required areas would tend to zero.