ScaleLab Surrogate Configuration Model
Expert-calibrated surrogate model configuration for industrial process optimization. Defines surrogate architectures, acquisition functions, objectives, and constraints used by the ScaleLab optimization engine.
Model Details
| Property | Value |
|---|---|
| Type | Surrogate configuration (no weights training required) |
| Framework | scikit-learn + numpy/scipy GP |
| Default Surrogate | Gaussian Process (Matérn-5/2, ARD) |
| Acquisition | Expected Improvement, qParEGO, Constrained EI |
Surrogate Models
| Model | Level | Description |
|---|---|---|
| Polynomial RSM | 2 | Quadratic response surface |
| Gaussian Process | 3 | Matérn kernel with ARD |
| Random Forest | 3 | 100-tree ensemble |
| Gradient Boosting | 3 | Stage-wise additive model |
| Neural Network | 3 | 2-hidden-layer MLP |
| Bayesian Neural Network | 4 | MC dropout uncertainty |
Files
surrogate_config.json— Full surrogate and acquisition configurationconfig.json— Model metadata
Usage
from huggingface_hub import hf_hub_download
import json
path = hf_hub_download("alirezaaminzadeh/scalelab-surrogate-config", "surrogate_config.json")
with open(path) as f:
config = json.load(f)
print(config["default_surrogate"])
Related Resources
- Space: experiment-to-scale-industrial-optimization-lab
- Instances: scalelab-benchmark-instances
- Benchmark: scalelab-benchmark-results