[Work Log] Testing CL energy

December 24, 2013

Testing constant-length energy function. First test: \(\eta\) and it's Jacobian, implemented in test/test_eta_deriv.m. Test passes. Interesting observation: jacobian is nearly bidiagonal. Hopefully the hessian will have similar form, so ignoring the off-diagonal terms won't be too detrimental.

Need to implement end-to-end test for \(E\) and its gradient/hessian. Compare against analytical Hessian estimate and the crude \(J'J\) Hessian approximator.

need to update likelihood means?

Posted by Kyle Simek
blog comments powered by Disqus