Closed TTitscher closed 3 years ago
As far as I see the changes in "vb.py", I think everything should be correct. And I believe, all the tests are also already adjusted accordingly. After merging the commit, I'll also test it with my own GDM problem.
When we ignore all that
k(theta) = (model(theta) - data)
vsk(theta) = (data - model(theta))
stuff, the Chapell paper basically defines the Jacobian J asJ = - dk / dtheta
, which can be confusing. Also it was documented incorrectly in our code.This PR changes this definition to
J = dk / dtheta
.