hydromodpy.calibration.adapters.gp_mapping_adapter#
Gaussian-process surrogate adapter with Expected Improvement acquisition.
The adapter works in the transformed parameter space exposed by
ParameterSpace. An initial
Latin-hypercube design is sampled and evaluated; subsequent iterations
fit a sklearn.gaussian_process.GaussianProcessRegressor
surrogate to the accumulated (x, y) pairs and pick the next point by
maximising Expected Improvement against the best observed value.
The optimizer declares convergence when EI at its argmax falls below a
configurable threshold (default 1e-6).
Classes
|
Gaussian-process surrogate optimizer using Expected Improvement. |