2 ms·

I've just found this for the exact same reason, for 2-10 sets. The gradient descent is done client-side (Tensorflow.js): https://www.deepvenn.com/ https://www.

• pimlottc 2y ago
  I've used this before, it's a cool idea but it only sort of works. For most non-trivial examples, there will be still be a big discrepancy between the visual area of overlaps and the and the actual element count. Claiming that the graph is area-proportional might actually be more misleading if it's not actually accurate. To improve it, you'd need more ways to mutate the shapes to try to find a solution with less error. E.g. if the circles could be squashed into ellipses, that would probably help. Of course, some higher-order graphs can't be shown with convex shapes at all, so you'd ideally want to be able to pull and stretch areas as needed. It's a hard problem.