Implementation of Active CMAES #367
Conversation
|
It's just the (Vanilla) CMAES tests that are failing. I'm fine tuning the parameters used in the tests. |
| #include <ensmallen_bits/function.hpp> | ||
|
|
||
| #ifndef NOT_EMPTY_TRANSFORMATION | ||
| #define NOT_EMPTY_TRANSFORMATION |
There was a problem hiding this comment.
not sure what this define is for
There was a problem hiding this comment.
Thank you for pointing this out! Now that I look at it, I don't think this define is required. The ACTIVE_CMAES_CMAES_IMPL_HPP define should be enough. I'll make changes.
There was a problem hiding this comment.
I'm looking at this once more and I realize that if the define is removed and the user uses both Vanilla and Active CMAES, there will be double inclusion (which is why I added the include guard in the first place). The ACTIVE_CMAES_CMAES_IMPL_HPP define is indeed not enough.
There was a problem hiding this comment.
It sounds like in that case that either NotEmptyTransfornation should be moved to its own file and included by both versions, or if the functionality differs here then perhaps it just needs a new name. I think I have understood the problem, let me know if not.
There was a problem hiding this comment.
Alright, I'll move NotEmptyTransfornation to its own file.
| @@ -20,31 +20,82 @@ | |||
|
|
|||
| #include <ensmallen_bits/function.hpp> | |||
|
|
|||
| #ifndef NOT_EMPTY_TRANSFORMATION | |||
| #define NOT_EMPTY_TRANSFORMATION | |||
There was a problem hiding this comment.
not sure what this define is for
There was a problem hiding this comment.
I'll be sure to update this too.
|
@mlpack-jenkins test this please |
|
If we use |
Yes, but the default is |
|
Do you see any problem with using |
|
We could even provide another constructor for thee empty transformation. |
|
Looking at it again, if we keep the constructor, it's backward compatible, since I can still use the "old" constructor. |
![mlpack-bot[bot]](https://avatars.githubusercontent.com/in/23934?s=60&v=4){kind=small}
Yes, that's why it was marked as deprecated instead of being removed altogether. We could make |
|
I think both interface are fine, the old interface is closer to what we provide for the other optimizers, so I like to keep it, the new one offers the ability to support more transformations so I like to keep it as well. |
|
Alright. We should probably modify Vanilla CMA-ES as well then (the element types should be changed as well for consistency). |
|
Makes sense, I'll resolve the failing test and merge this PR. |
|
Thanks for the great contribution. |
This pull request is an initial implementation of Active CMAES, which is part of my Google Summer of Code (GSoC) project. Tests have not been added yet, but they will be included soon.