Support for constrained optimization?
ogencoglu opened this issue · 2 comments
Any plans on supporting arbitrary constraints on input/output?
E.g. non-linear inequality constraints etc.
Thanks for creating this issue. We will have a discussion regarding the suggestion soon, and hopefully, we might extend Mango with capabilities to support arbitrary constraints as well.
Dear Oguzhan Gencoglu,
We will be able to support the functionality of constraints specification in Mango in the future release. We believe this will be a good contribution to extend to more use-cases. This will take us some time to think through the required syntax changes and accordingly the changes in internal search implementation as well. We know our current search implementation can support it. It is a matter of extending the search space syntax so users can specify even richer constraints.
In the current syntax, the search domain is kind of the constraint within which the user is interested to find a solution. But our current search domain specification cannot support arbitrary constraints on input/output. The search domain syntax is mostly influenced by the ML parameter search space.
regards,
Sandeep Singh.