[documentation] Update the performance tips #328
Conversation
|
There was a problem hiding this comment.
@amontoison nice. regarding the syntax, it's getting a bit heavy. though it does not seem to be standard in Julia yet, I'd advocate sth like what used to be done, e.g., in matlab: build iteratively the options using an ad hoc constructor, then, pass it to the code. or having a dedicated keyword for such "empty" backends...
There was a problem hiding this comment.
@tmigot Any idea of how we can design that with the current API?
There was a problem hiding this comment.
@jbcaillau do you mean having something like:
ADNLPModel(f, x0, c, lcon, ucon, gradient_backend)
ADNLPModel(f, x0, c, lcon, ucon, gradient_backend, jacobian_backend)
ADNLPModel(f, x0, c, lcon, ucon, gradient_backend, jacobian_backend, jprod_backend, jtprod_backend)
ADNLPModel(f, x0, c, lcon, ucon, gradient_backend, jacobian_backend, jprod_backend, jtprod_backend, hessian_backend)
...
I find it kind of heavy and it will not be 100% clear or do you mean something different ?
Note that in the code right now, all these backends objects all are subtypes of ADBackend, so we don't have ways to distinguish between different types of backends without keywords.
@jbcaillau