Example on vector CostTerm ? about ifopt HOT 4 CLOSED

Well this does not fully answer my question but I shall dig in the code a little more and come back to you with a clearer idea.
Thanks a lot anyway !

from ifopt.

Hi!
In the end, the cost always has to be scalar, that's just the way nonlinear programming problems are defined. This cost however can be composed of a variety of scalar cost-terms. You can subtract a scalar cost term by defining the costs as negative.

The image shows the scalar cost J(x), and the gradient of this cost w.r.t x, which are two added 1xn row vectors (n being the number of optimization variables). Does that clear things up?

from ifopt.

Thanks for the (very) quick answer !

From a theoretical point of view we're on the same page. My question however was more from a programming point of view.
Using the same example as before, say I would like to do this vector subtraction (or whatever the operation) all at once. To benefit from vectorization, because the operation is heavy, well whatever the reason... How could one somewhat 'pre-compute' the error (in its vector form) to later on 'distribute' each scalar cost-term ? Or how to 'group' this variety of CostTerm objects so that the operation is computed only once ?
I'm not sure this comment make things clearer, hopefully.

from ifopt.

So I guess I would simply implement this as one cost term, and then exploit Eigen efficiency to calculate the cost e.g. as

VectorXd x0 = var_set0->GetValues();
VectorXd x1 = var_set1->GetValues();
double cost = (x0 - x1).norm(); // Eigen will do this in the fastest way possible I assume

from ifopt.