You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
POEM 096 proposes the ability to minimize the constraint violation from some starting point of driver. This would allow users to find a feasible starting point given the current design variable values.
SNOPT supports the option to find a feasible starting point, but this capability isn't available in other optimizers.
Adding it to OpenMDAO would be a valuable addition.
The POEM proposes a capability, but not an implementation, and that needs to be developed.
In this case, we're asking the optimizer to find the feasible starting point, so it seems like it should be associated with the driver.
Proposal for initial implementation:
Calling prob.run_driver(find_feasible=True) will replace the user-defined objective with a new objective that is the norm of the constraint violations.
This new objective is not the output of an OpenMDAO component itself, so we will have to implement the reverse mode derivatives more manually.
Associated POEM
096
The text was updated successfully, but these errors were encountered:
is it possible to 'claim' (ie., be assigned) this issue? A partner and I are working on attempting to make a contribution to an open source project for one of our classes and want to take a stab at this problem.
By all means, we'd be happy to accept this if you'd like to undertake it. Post an issue on github and then assign it to yourself. (If you find yourself unable to self-assign it let me know and I'll work it out).
Thank you! We recognize that this problem is complex and our solution may not be accepted, but we thought it would be interesting to make an attempt at this problem because of it's complexities
Desired capability or behavior.
POEM 096 proposes the ability to minimize the constraint violation from some starting point of driver. This would allow users to find a feasible starting point given the current design variable values.
SNOPT supports the option to find a feasible starting point, but this capability isn't available in other optimizers.
Adding it to OpenMDAO would be a valuable addition.
The POEM proposes a capability, but not an implementation, and that needs to be developed.
In this case, we're asking the optimizer to find the feasible starting point, so it seems like it should be associated with the driver.
Proposal for initial implementation:
Calling
prob.run_driver(find_feasible=True)
will replace the user-defined objective with a new objective that is the norm of the constraint violations.This new objective is not the output of an OpenMDAO component itself, so we will have to implement the reverse mode derivatives more manually.
Associated POEM
096
The text was updated successfully, but these errors were encountered: