About SKSurrogate

Software you can audit, not just install: an open-source, scikit-learn-native environment for optimizing expensive objectives, where every number it reports is traceable to a protocol, a seed and an archived record.

What it is

SKSurrogate implements surrogate-guided optimization for objectives that are expensive to evaluate: each evaluation costs minutes rather than milliseconds, so the question is not how many candidates you can try but how much you learn from each one. The package builds a response surface over the space you have sampled, then uses that surface to decide where to spend the next evaluation.

Around that core it provides the machinery a research or production workflow needs: dataset registration with identity, a drop-in BaseSearchCV tuner, statistical evaluation with nested cross-validation and conformal intervals, sensitivity and dependence analysis, experiment design for expensive runs, a control plane that executes and logs searches, and a research workspace that keeps each claim attached to the evidence that supports it.

What it is not

Not a managed cloud service. Self-hosted and single-machine by design — local API or one compose profile. If you need multi-node training or GPU fleets, that is a different category of tool.

No LLM magic in the box. The shipped tooling is deterministic. A bounded, opt-in cited-note synthesis (your endpoint, aggregate facts only, enforced citations) already runs inside it; a conversational research assistant is roadmap, not this release.

Not a black-box AutoML zoo. The search space is your pipeline plus the ranges you declare, and every evaluation is recorded — results are auditable artifacts, not opaque picks from a fixed model zoo.

What is dated and gated lives on the roadmap.

How it is developed

Development happens in the open on GitHub. The benchmark suite is part of the repository, not a marketing appendix: a frozen protocol contract, pre-registered hypotheses with explicit falsifiers, and raw run records archived with DOIs so a reader can check the advertised results without rerunning anything.

That has two consequences. The changelog is public and versioned, documentation lives at a stable address on Read the Docs instead of being regenerated per release, and when the data says the package does not beat a baseline, that gets published too — see the honest benchmark post.

Who maintains it

SKSurrogate is written and maintained by Mehdi Ghasemi, a scientist-mathematician. Contact is by email or through the issue tracker — see contact. There is no support contract and no service-level agreement attached to the project; it is maintained software, not a hosted product.

GitHub ↗ · Google Scholar ↗ · ORCID ↗ · mghasemi.net ↗

Licence

SKSurrogate is released under the MIT licence. You may use it commercially, modify it and redistribute it, provided the copyright notice and permission notice are retained. The canonical wording is the MIT licence.