Reinforcement Learning Toolbox 2.0 last updated:

General Documentation Manual Tutorial Class Reference Master Thesis Examples Related Papers Downloads Links News mailto:webmaster Main Page     Class Hierarchy   Compound List   File List   Compound Members   File Members CAdaptiveParameterFromAverageRewardCalculator Class Reference Adaptive Parameter Calculator which calculates the parameter's value from the current average reward. More... #include <cagentlistener.h> Inheritance diagram for CAdaptiveParameterFromAverageRewardCalculator: List of all members. Public Member Functions   CAdaptiveParameterFromAverageRewardCalculator (CParameters *targetObject, string targetParameter, CRewardFunction *reward, int nStepsPerUpdate, int functionKind, double paramMin, double paramMax, double targetMin, double targetMax, double alpha)   ~CAdaptiveParameterFromAverageRewardCalculator () virtual void  nextStep (CStateCollection *oldState, CAction *action, double reward, CStateCollection *newState)   virtual function, to be implemented by subclass virtual void  onParametersChanged ()   Updates all data elements represents parameters. virtual void  resetCalculator ()   Reset the targetValue. Protected Attributes double  alpha double  targetValue int  nSteps int  nStepsPerUpdate Detailed Description Adaptive Parameter Calculator which calculates the parameter's value from the current average reward. The target value in this class is the current average reward. The target value gets resetted the minimim expected reward if a new learning trial has started. This adaptive parameter has to be added to the agent's listener list in order to calculate the average reward. The average reward is calculated dynamically with the formular averagereward_t+1 = averagereward_t * alpha + reward_t+1 * (1 - alpha). Alpha can be set with the parameter "APRewardUpdateRate" and defines the update rate of the average reward. Alpha should be choosen close to 0.99 to get good results. The average reward is not resetted when a new episode begins. For more details see the super class. Parameters of CAdaptiveParameterFromNStepsCalculator: "APRewardUpdateRate": Update rate for the average reward. see CAdaptiveParameterBoundedValuesCalculator Constructor & Destructor Documentation CAdaptiveParameterFromAverageRewardCalculator::CAdaptiveParameterFromAverageRewardCalculator (  CParameters *  targetObject, string  targetParameter, CRewardFunction *  reward, int  nStepsPerUpdate, int  functionKind, double  paramMin, double  paramMax, double  targetMin, double  targetMax, double  alpha )  CAdaptiveParameterFromAverageRewardCalculator::~CAdaptiveParameterFromAverageRewardCalculator (   )  Member Function Documentation virtual void CAdaptiveParameterFromAverageRewardCalculator::nextStep (  CStateCollection *  oldState, CAction *  action, double  reward, CStateCollection *  newState )  [virtual]   virtual function, to be implemented by subclass Reimplemented from CSemiMDPRewardListener. virtual void CAdaptiveParameterFromAverageRewardCalculator::onParametersChanged (   )  [virtual]   Updates all data elements represents parameters. Reimplemented from CAdaptiveParameterBoundedValuesCalculator. virtual void CAdaptiveParameterFromAverageRewardCalculator::resetCalculator (   )  [virtual]   Reset the targetValue. This function is used for resetting for example the steps or number of episodes when learning is restarted. (used for parameter evaluation) Implements CAdaptiveParameterCalculator. Member Data Documentation double CAdaptiveParameterFromAverageRewardCalculator::alpha [protected] int CAdaptiveParameterFromAverageRewardCalculator::nSteps [protected] int CAdaptiveParameterFromAverageRewardCalculator::nStepsPerUpdate [protected] double CAdaptiveParameterFromAverageRewardCalculator::targetValue [protected] The documentation for this class was generated from the following file: cagentlistener.h