pestpp-glm:: defining custom loss (objective) function

[jbensabat]

Hello
I am running pestpp-glm and so far using the weight sum of the residual squares as objective (loss) function, the one implemented in the code. I would like to test alternative loss functions. Is there a way to do so without amending the code ?
best
jac

[jtwhite79]

People typically do this by adding those other loss/objective functions as "observations" in the control file and then (try to) seek a preferred value for them, otherwise, some pretty serious code mods are required. I believe the reason we use the sum-of-squared weighted residuals is because it is coherent with the assumptions of Gauss's approximation to the Hessian in Newton's equation, but I could be wrong about that...

[jbensabat]

hi
thanks for your reply. This means that pestpp implements only sum of squares loss function (it can be the residual or a function of the residual). I will look into the code to see what it takes to allow additional loss functions
best
jac