Hi!
Sorry for the delay in replying. The reason is that we switched to a new optimization-related communication protocol that can do nice tricks like parallel numerical differentiation, but has difficulty supporting Hessians. We were also under impression that this part of ALGLIB was almost unused, because almost anyone who contacted us used gradient/Jacobian-based optimizers.
Finally, I feel that Hessian-based optimization should not be a part of LSFIT solver, it is a legacy of questionable decision made when developing the very first versions of ALGLIB.
Sorry for any inconvenience caused by this backward-incompatible change.
Sergey
|