ribs.emitters.opt

Various optimizers which are employed across emitters.

There are gradient-based optimizers which inherit from GradientOptBase and evolution strategies which inherit from EvolutionStrategyBase. When specifying optimizers for an emitter, one can pass in the optimizer class itself, or the string name of the optimizer, or an abbreviated name. The supported abbreviations are as follows.

For gradient-based optimizers (inheriting from GradientOptBase):

• adam: AdamOpt

• gradient_ascent: GradientAscentOpt

For evolution strategies (inheriting from EvolutionStrategyBase):

• cma_es: CMAEvolutionStrategy

• sep_cma_es: SeparableCMAEvolutionStrategy

• lm_ma_es: LMMAEvolutionStrategy

• openai_es: OpenAIEvolutionStrategy

ribs.emitters.opt.CMAEvolutionStrategy(...)	CMA-ES optimizer for use with emitters.
ribs.emitters.opt.LMMAEvolutionStrategy(...)	LM-MA-ES optimizer for use with emitters.
ribs.emitters.opt.OpenAIEvolutionStrategy(...)	OpenAI-ES optimizer for use with emitters.
ribs.emitters.opt.SeparableCMAEvolutionStrategy(...)	sep-CMA-ES optimizer for use with emitters.
ribs.emitters.opt.EvolutionStrategyBase(...)	Base class for evolution strategy optimizers for use with emitters.
ribs.emitters.opt.AdamOpt(theta0[, lr, ...])	Adam optimizer.
ribs.emitters.opt.GradientAscentOpt(theta0, lr)	Vanilla gradient ascent.
ribs.emitters.opt.GradientOptBase(theta0, lr)	Base class for gradient-based optimizers.