Breaking gradients with fabs and the "double monomial"

[rok_cesnovar]

@bnicenboim

which unless I’m very wrong fabs is fine

fabs(x) is not differentiable at x = 0

A bit more on that here: Sqrt(square()) vs. fabs()