Add the direct numeric calculation of the likelihood gradient (not finite differences). All of our dials are continuous, so they can have the gradient calculated. Propagating the dial gradients through to the likelihood will need a bit of work, but looks possible. I've attached a summary of the math that we need.
GUNDAM_Gradients.pdf
(I have the tex source for when we find mistakes in the math)
Add the direct numeric calculation of the likelihood gradient (not finite differences). All of our dials are continuous, so they can have the gradient calculated. Propagating the dial gradients through to the likelihood will need a bit of work, but looks possible. I've attached a summary of the math that we need.
GUNDAM_Gradients.pdf
(I have the tex source for when we find mistakes in the math)