4 ms·

> CG code to calculate area proportional venn diagrams Is the matrix in this case guaranteed to be positive definite? I would expect a non-trivial nullspace, a

• benfrederickson 10y ago
  For the venn code, I wrote up the approach here : http://www.benfrederickson.com/better-venn-diagrams/ http://www.benfrederickson.com/better-venn-diagrams/ Basically though, I'm using the non-linear CG method - so it doesn't require a positive definite matrix. The loss function is a little funky with handling the disjoint set/ subset relationships in the euler diagrams appropriately (defines the loss/gradient to be 0 if these constraints are satisfied), but this approach still works pretty well. That venn diagram post has a couple interactive demos of how this works, and also a randomized test showing overall performance. I actually believe its the best known algorithm for laying out area proportional venn diagrams. I benchmarked against the code from the venneuler paper here: http://benfred.github.io/venn.js/tests/venneuler_comparison/ http://benfred.github.io/venn.js/tests/venneuler_comparison/