surrogate

surrogate

Surrogate models for SpotOptim.

This module provides two Kriging (Gaussian Process) implementations:

1. Kriging: Full-featured implementation with:
   • Multiple methods: interpolation, regression, reinterpolation
   • Mixed variable types: float/num, int, factor
   • Isotropic/anisotropic correlation
   • Lambda (nugget) optimization for regression
   • Compatible with SpotOptim’s variable type conventions
2. SimpleKriging: Lightweight implementation with:
   • Gaussian kernel only
   • Basic hyperparameter optimization
   • Faster for simple problems
   • Limited to continuous variables

For most SpotOptim applications, use Kriging (the default). Use SimpleKriging for quick prototyping or simple continuous problems.

Examples

>>> from spotoptim.surrogate import Kriging
>>> model = Kriging(method='regression', seed=42)
>>> model.fit(X_train, y_train)
>>> predictions = model.predict(X_test)

>>> from spotoptim.surrogate import SimpleKriging
>>> simple_model = SimpleKriging(noise=1e-10, seed=42)
>>> simple_model.fit(X_train, y_train)