alirezaaminzadeh/scalelab-benchmark-results

ScaleLab Benchmark Results

Reproducible benchmark results comparing optimization algorithms across industrial process instances.

Benchmark Schema

Each result record contains:

Field	Description
instance_id	Instance identifier
algorithm	Optimization algorithm used
solver	Surrogate/solver backend
configuration	Algorithm configuration
seed	Random seed
runtime_sec	Total runtime
time_to_first_solution	Time to reach target quality
n_experiments	Total experiments conducted
best_objective	Best observed objective value
optimality_gap	Gap to estimated optimum (%)
constraint_violation	Total constraint violation
hypervolume	Multi-objective hypervolume
surrogate_calibration	Surrogate R² calibration score
robustness_score	Scale-up robustness score
scaleup_performance	Plant performance retention
control_tracking_error	MPC tracking RMSE
energy_consumption	Energy at optimum (kWh)
status	optimal / feasible / infeasible

Algorithms Compared

• Grid Search (Level 1)
• Random Search (Level 1)
• Latin Hypercube + GP-BO (Level 3)
• Constrained BO (Level 3)
• Full Pipeline (Level 5)

Usage

import json
from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="alirezaaminzadeh/scalelab-benchmark-results",
    filename="eval_results.json",
    repo_type="dataset",
)
with open(path) as f:
    results = json.load(f)
print(results["leaderboard"])

Related Resources

• Space: experiment-to-scale-industrial-optimization-lab
• Instances: scalelab-benchmark-instances