Make theta a user-supplied array param #370
Conversation
Codecov Report
@@ Coverage Diff @@
## master #370 +/- ##
==========================================
+ Coverage 78.76% 78.97% +0.21%
==========================================
Files 16 16
Lines 2270 2269 -1
==========================================
+ Hits 1788 1792 +4
+ Misses 482 477 -5
📣 Codecov can now indicate which changes are the most critical in Pull Requests. Learn more |
vikram-s-narayan
changed the title
|
Is the plan to add a gradient free optimization method for the user for now while you work on #371 ? |
| @@ -35,15 +35,14 @@ function bounds_error(x, xl) | |||
| end | |||
|
|
|||
| #constructor for GEKPLS Struct | |||
| function GEKPLS(X, y, grads, n_comp, delta_x, xlimits, extra_points, θ) | |||
| function GEKPLS(X, y, grads, n_comp, delta_x, xlimits, extra_points, theta) | |||
There was a problem hiding this comment.
We should docstring these. Anyways, that's for the future.
|
I'm not sure the hyperparameter optimization needs to be in the library with this feature. You just create and run an OptimizationProblem tuning theta and it should be fine. |
Users can use BlackBoxOptim.jl or equivalent on their own (outside of the GEKPLS system) until we make GEKPLS differentiable. After that, users will still be doing hyperparameter optimization on their own outside of the system with their choice of gradient-based optimization |
|
Set it up to make an Optimization.jl OptimizationProblem and it'll then have access to all optimizers. Then it will just fail if you try to |
Addresses #366 by enabling users to supply their own theta array as a param. This will allow users to do hyperparameter optimization outside of the GEKPLS system.