Skip to content

2D: set a better default loss

Bas Nijholt requested to merge default_loss into master

Currently when running the notebook on a computer with 4 cores or less the Learner2D example in learner.ipynb will stop after the bounds are done.

Adding the area to the loss will ensure there is a minimal area reached.

Personally, I've been using this loss dev * np.sqrt(A) + 0.3 * A (I added a small prefactor, but that will probably be more difficult to explain) for a while now and it works quite well for me.

Merge request reports