Policy Search Methods

Interface for MMLF policy search methods

MMLF policy search methods must implement the following methods:

• getActivePolicy
• getBestPolicy
• tellFitness

class resources.policy_search.policy_search.PolicySearch

Interface for MMLF policy search methods

MMLF optimizers must implement the following methods:

• getActivePolicy(): Returns the currently active policy

• getBestPolicy(): Returns the best policy found so far

• tellFitness(fitnessSample): Gives one fitness sample for the currently

  active policy

static create(policySearchSpec, stateSpace, actionSpace, stateDims)

Factory method that creates policy search method based on spec.

getActivePolicy()

Returns the currently active policy

getBestPolicy()

Returns the best policy found so far

static getPolicySearchDict()

Returns dict that contains a mapping from PS name to PS class.

reset()

Resets the policy search method

tellFitness(fitnessSample)

Gives one fitness sample for the currently active policy

Provides the fitness fitnessSample obtained in one evaluation of the currently active policy.

Fixed Parametrization

Policy search method that optimizes the parameters of a fixed policy class

The module contains a policy search method that optimizes the parameters of an externally predefined, parametrized policy class. It can use an arbitrary black-box optimization algorithm for this purpose.

class resources.policy_search.fixed_parametrization.FixedParametrization(policy, optimizer)

Policy search method that optimizes the parameters of a policy

The module contains a policy search method that optimizes the parameters of an externally predefined policy.

CONFIG DICT

policy:: A parametrized class of policies optimizer:: An arbitrary black-box optimizer