First light!
- Bayesian optimization via
gp_minimize
. - Tree-based sequential model-based optimization via
forest_minimize
andgbrt_minimize
, with support for multi-threading. - Support of LCB, EI and PI as acquisition functions.
- Plotting functions for inspecting convergence, evaluations and the objective function.
- API for specifying and sampling from a parameter space.
- Speed improvements when using
gp_minimize
withacq_optimizer='lbfgs'
andacq_optimizer='auto'
when all the search-space dimensions are Real. - Persistence of minimization results using
skopt.dump
andskopt.load
. - Support for using arbitrary estimators that implement a
return_std
argument in theirpredict
method by means ofbase_minimize
fromskopt.optimizer.
- Support for tuning noise in
gp_minimize
using thenoise
argument. TimerCallback
inskopt.callbacks
to log the time between iterations of the minimization loop.
See AUTHORS.md
.