Currently our library offers two ways of searching values for parameterised nodes: Evolution Strategies and Gradient based. Both search values "locally" around the current values. In some use cases one might not want to search locally, but, e.g. over values of different magnitude. One possible way to do this is by randomly sampling possible values for each parameter and updating parameters when fitness increases. Such a search should be added to the local search implementations.
Currently our library offers two ways of searching values for parameterised nodes:
Evolution StrategiesandGradient based. Both search values "locally" around the current values. In some use cases one might not want to search locally, but, e.g. over values of different magnitude. One possible way to do this is by randomly sampling possible values for each parameter and updating parameters when fitness increases. Such a search should be added to the local search implementations.