Skip to content

README.md

Latest commit

 

History

History
21 lines (16 loc) · 705 Bytes

README.md

File metadata and controls

21 lines (16 loc) · 705 Bytes
Raw

Optimize the optimizer

EvoX²-style benchmark: the discovery object is a search controller (selection + mutation / ask policy), scored by how well it optimizes a portfolio of black-box functions under a fixed eval budget.

Held-out problems (Ackley, Rosenbrock) dominate combined_score so the controller must transfer, not memorize the train suite.

Run

uv run skydiscover-run benchmarks/optimize_the_optimizer/initial_program.py \
  benchmarks/optimize_the_optimizer/evaluator.py \
  -c benchmarks/optimize_the_optimizer/config.yaml -s best_of_n -i 50
python3 benchmarks/optimize_the_optimizer/evaluator.py \
  benchmarks/optimize_the_optimizer/initial_program.py