| name | egobox-egor-tuning |
| description | Use this skill whenever the user is working with the Egor optimizer. Triggers on any mention of: Egor optimizer Bayesian optimization in Rust/Python or requests to minimize expensive black-box functions using surrogate models.
|
EGOR Tuning Skill
Goal
Automatically select and adapt EGOR optimization parameters based on:
- problem dimension
- objective evaluation runtime
- available optimization budget
- parallel evaluation availability
- convergence progress
Workflow
- Estimate problem characteristics.
- Select initial EGOR strategy.
- Monitor optimization progress.
- Adapt exploration/exploitation.
Basic rules
- More dimensions -> reduce direct sampling, use surrogate strategies.
- Expensive evaluations -> maximize information gain.
- Poor progress -> increase exploration.
Advanced strategies
Poor convergence
If optimization stagnates:
- enable TREGO
- switch kernel from SquaredExponential to Matern52
- increase exploration
Dimension > 10
Use KPLS for Gaussian process fitting.
Examples:
- dimension 20: kpls_dim=5
- dimension 100: kpls_dim=10
Dimension > 50
Use CoEGO.
Example:
- dimension 100: n_coop_comp=5
Parallel evaluations
If evaluations can run in parallel:
use qEIConfig.
Examples:
- dimension 50: batch=5
- dimension 100: batch=10
