Open yannikschaelte opened 6 years ago
keep in mind we then need to calculate and use the derivative of J with respect to sigma, since it is not 0 anymore
We have already computed the derivatives dJ/dsigma, dJ/dc etc., so this should theoretically be no problem, and practically at least for forward sensitivities not too difficult to implement. But it might be not the most urgent issue ...
Sometimes, one wants to have hierarchically optimized parameters (e.g. sigmas), but also confine them to some box constraints [lb ub]. Therefore, having computed the optimal parameters in the inner loop, check if they stay in the box (if one was passed), and set them to the left or right boundary if not.