search

numericalalgorithmsgroup / pybobyqa

Python-based Derivative-Free Optimization with Bound Constraints
https://numericalalgorithmsgroup.github.io/pybobyqa/
GNU General Public License v3.0
78 stars 18 forks source link

Hessian class #8

Closed lindonroberts closed 4 years ago

lindonroberts commented 5 years ago

Speed up trust region solve by implementing Hessian-vector multiplication in C (e.g. with Cython).

lindonroberts commented 5 years ago

Looking online, consensus seems to be that just using a full square matrix is better for symmetric matrices, since LAPACK assumes this anyway. So, just use a normal np.array instead of Hessian class.

e.g. https://stackoverflow.com/a/9455669

lindonroberts commented 4 years ago

Fixed in v1.2 (using np.ndarray for Hessian throughout)