Add Michalewicz benchmark #464
Conversation
|
For the current number of DOE iterations and MC runs, I obtained the following result which looks like a good baseline to me: |
| ) | ||
|
|
||
|
|
||
| def synthetic_michalewicz(settings: ConvergenceExperimentSettings) -> DataFrame: |
There was a problem hiding this comment.
add a 5d to the name to make the dimensionality clear?
There was a problem hiding this comment.
can we add the global optimum as dashed line to the simulation plot you showed? Results without that can be deceiving
There was a problem hiding this comment.
That should be the case when the code is being run in the dashboard as you can see here for the current benchmark https://baybe-benchmark.apps.p.uptimize.merckgroup.com/ (@fabianliebig please confirm)
The plot I posted here was just meant as a verification that the chosen DoE and MC numbers are somewhat reasonable.
| X1, X2 = np.meshgrid(X1, X2) | ||
|
|
||
| # Michalewicz function | ||
| Z = -1 * ( |
There was a problem hiding this comment.
feel to me you should reuse the 5d variant shown above, just lot a 2d slice (or several)
reason: if theres a problem int he function above this plot wouldnt reveal it
There was a problem hiding this comment.
I will try my best :D
This PR adds a new benchmark, the 5-dimensional variant of the Michalewicz function.
Details on this function can be found here. The implementation as well as the choice of the number of dimensions is based on the previous benchmark that we had on this.