SKSurrogate in five minutes

Install

pip install sksurrogate

The package requires Python 3.10 or newer and ships the search layer on top of scikit-learn. Nothing else is needed for the tuner; the control plane runs from a compose profile when you want it.

Optimize a model

from sklearn.svm import SVC
from SKSurrogate import SurrogateRandomCV

search = SurrogateRandomCV(
    SVC(),
    param_distributions={"C": (1e-3, 10.0), "kernel": ["poly", "rbf"], "degree": (1, 4)},
    n_iter=30,
).fit(X, y)
search.best_params_, search.best_score_

The object is a BaseSearchCV subclass, so best_params_, best_score_, cv_results_, refitting and scoring behave as they do in scikit-learn — adopting it changes one import, not your codebase. Every candidate evaluation is recorded, so the result is an artifact you can inspect rather than a single returned winner.

Automatic pipeline design

from SKSurrogate import AML

AML(preset="classification").fit(X, y)

AML(preset="regression") covers the regression case and needs no search configuration. For pandas DataFrames the preset fits numeric, categorical, datetime and text preprocessing inside each cross-validation fold when cat_cols is omitted; array-like inputs should already be numeric. fit(X, y, strategy="eoa") selects evolutionary search.

Where to go next

• Surrogate search guide ↗ — strategies, acquisition, and how the response surface is fitted.
• Preprocessing guide ↗ — how presets build fold-local preprocessing.
• Control plane ↗ — the API, the UI and the compose profile on port 8013.
• Benchmarks — what the search claims, and where it does not win.
• Concepts — what surrogate-based optimization is, and when it pays.