-
Notifications
You must be signed in to change notification settings - Fork 11
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add BFGS #92
base: devel
Are you sure you want to change the base?
Add BFGS #92
Conversation
- a class to estimate (inverse) Hessian from iterates + grad
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Good work in progress @fabinsch.
I made some minor comments related to the coding style ;)
Co-authored-by: Justin Carpentier <[email protected]>
- proof of concept to show that bfgs is working within the proxnlp solver to estimate hessians - solve the ur5-ik task with BFGS (86iters) and compare against exact hessian (19iters) and identity hessian (345iters)
For now, the As a next step, |
Let's schedule some time after Nov. 15th to work on this again @fabinsch. |
Start of work to add the Broyden–Fletcher–Goldfarb–Shanno algorithm. For now, it is a class that estimates an (inverse) Hessian from iterates and the corresponding gradients at these iterates. An optimizer can be built around it.
It implements the BFGS update rule (eqn 6.17 p. 140) to estimate the inverse Hessian and the DFP update rule (eqn 6.13 p.139) from Nocedal and Wright, Numerical Optimization, 2nd edition, which have a similar structure.