Nonlinear SDP interior point meta-algorithm #106
Conversation
Co-authored-by: Mathieu Besançon <mathieu.besancon@gmail.com>
|
Refactored following advised by @mohamed82008 . But still debugging the StackOverflow Error. |
|
Notifying @matbesancon as well |
Co-authored-by: Mohamed Tarek <mohamed82008@gmail.com>
|
Once the minor comments above are addressed and the tests pass, please add documentation on this. Then I am ready to merge this. |
|
let's have a separate PR for docs. |
|
thanks @noilreed, great PR 🎉 |
| @@ -100,7 +100,7 @@ end | |||
| function sd_objective(objective0, sd_constraints, c::AbstractArray) | |||
| function _objective(args) | |||
| target = objective0(args) | |||
| barrier = sum(c .* -logdet.(sd_constraints.fs.(args))) | |||
| barrier = sum(c .* -logdet.(map(f -> f(args), sd_constraints.fs))) | |||
There was a problem hiding this comment.
@mohamed82008 What makes this different? I found weird thing in my local version test yesterday before went to bed and I haven't pushed it, not sure if that's a bug. That if two MvNormal applied to one optimization then the first one is not well optimized, is this the reason?
There was a problem hiding this comment.
Yes, broadcasting a function calls it element-wise on the arguments. We need to call each function with all the arguments as input, hence the map.
No description provided.