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 CGradientUpdateFunction Class Reference Interface class for all function which gradient update. More... #include <cgradientfunction.h> Inheritance diagram for CGradientUpdateFunction: List of all members. Public Member Functions   CGradientUpdateFunction () virtual  ~CGradientUpdateFunction () void  updateGradient (CFeatureList *gradientFeatures, double factor=1.0)   Does the preprocessing for the gradient update. virtual void  updateWeights (CFeatureList *dParams)=0   Interface for updating the weights. virtual int  getNumWeights ()=0   Returns the number of weights. virtual CAdaptiveEtaCalculator *  getEtaCalculator ()   get the eta calculator for this gradient update function virtual void  setEtaCalculator (CAdaptiveEtaCalculator *etaCalc)   set the eta calculator for this gradient update function virtual void  getWeights (double *parameters)=0   Function for getting all weights. virtual void  setWeights (double *parameters)=0   Function for setting all weights. virtual void  saveData (FILE *stream)   Save weights coming from getWeights. virtual void  loadData (FILE *stream)   Load weights and set them with setWeights. virtual void  resetData ()=0   Interface for resetting the weights. virtual void  copy (CLearnDataObject *gradientFuntion) Protected Attributes CFeatureList *  localGradientFeatureBuffer CAdaptiveEtaCalculator *  etaCalc Detailed Description Interface class for all function which gradient update. Gradient update function only support updating the weights of the function, given the gradient. The datastructure of the gradient is always a feature list, because for RBF networks many parts of the gradient are zero. Additionally you can retrieve and set the weights directly with the functions getWeights(double *parameters) and setWeights(double *parameters). Gradient update functions don't implement the functions itself, so you can't calculate the output of the function given an input, you also can't calculate the gradient itself. This interface is just for the update of the weights! Each subclass of CGradientUpdateFunction has to implement the functoins: updateWeights(CFeatureList *gradient): Update the weights according to the gradient. getWeights(double *parameters), write all weights in the double array setWeights(double *parameters), set the weights according to the double array resetData(): reset all weights, needed when a new learning process is started getNumWeights(): return the number of weights. When the weights are updated, the function updateGradient(CFeatureList *gradientFeatures, double factor = 1.0) is called. The gradient is at first saved in the local gradient buffer and then multiplied by the specified factor. If an eta calculator exists the eta calulator is applied to the gradient update, after that the user-interface function updateWeights is called. Parameters of CGradientUpdateFunction: The gradient update function inherits all Parameters from its eta calculator. Constructor & Destructor Documentation CGradientUpdateFunction::CGradientUpdateFunction (   )  virtual CGradientUpdateFunction::~CGradientUpdateFunction (   )  [virtual] Member Function Documentation virtual void CGradientUpdateFunction::copy (  CLearnDataObject *  gradientFuntion  )  [virtual]   Reimplemented from CLearnDataObject. Reimplemented in CGradientQFunction, CQFunction, CFeatureVRegressionTreeFunction, and CGradientVFunction. virtual CAdaptiveEtaCalculator* CGradientUpdateFunction::getEtaCalculator (   )  [virtual]   get the eta calculator for this gradient update function virtual int CGradientUpdateFunction::getNumWeights (   )  [pure virtual]   Returns the number of weights. Implemented in CContinuousActionGradientPolicy, CContinuousActionPolicyFromGradientFunction, CContinuousActionFeaturePolicy, CContinuousActionSigmoidPolicy, CContinuousActionQFunction, CCALinearFAQFunction, CContinuousTimeAndActionSigmoidVMGradientPolicy, CQFunction, CComposedQFunction, CTorchGradientFunction, CVFunctionFromGradientFunction, CQFunctionFromGradientFunction, and CFeatureVFunction. virtual void CGradientUpdateFunction::getWeights (  double *  parameters  )  [pure virtual]   Function for getting all weights. The double array is assumed to be large enough. This isn't checked! Implemented in CContinuousActionGradientPolicy, CContinuousActionPolicyFromGradientFunction, CContinuousActionFeaturePolicy, CContinuousActionSigmoidPolicy, CContinuousActionQFunction, CCALinearFAQFunction, CContinuousTimeAndActionSigmoidVMGradientPolicy, CQFunction, CComposedQFunction, CTorchGradientFunction, CVFunctionFromGradientFunction, CQFunctionFromGradientFunction, and CFeatureVFunction. virtual void CGradientUpdateFunction::loadData (  FILE *  stream  )  [virtual]   Load weights and set them with setWeights. Implements CLearnDataObject. Reimplemented in CGradientQFunction, CQFunction, CComposedQFunction, CGradientVFunction, and CFeatureVFunction. virtual void CGradientUpdateFunction::resetData (   )  [pure virtual]   Interface for resetting the weights. Implements CLearnDataObject. Implemented in CContinuousActionGradientPolicy, CContinuousActionPolicyFromGradientFunction, CContinuousActionFeaturePolicy, CContinuousActionSigmoidPolicy, CContinuousTimeAndActionSigmoidVMGradientPolicy, CGradientQFunction, CQFunction, CComposedQFunction, CTorchGradientFunction, CVFunctionFromGradientFunction, CQFunctionFromGradientFunction, CRegressionTreeQFunction, CGradientVFunction, and CFeatureVFunction. virtual void CGradientUpdateFunction::saveData (  FILE *  stream  )  [virtual]   Save weights coming from getWeights. Implements CLearnDataObject. Reimplemented in CGradientQFunction, CQFunction, CComposedQFunction, CGradientVFunction, and CFeatureVFunction. virtual void CGradientUpdateFunction::setEtaCalculator (  CAdaptiveEtaCalculator *  etaCalc  )  [virtual]   set the eta calculator for this gradient update function virtual void CGradientUpdateFunction::setWeights (  double *  parameters  )  [pure virtual]   Function for setting all weights. The double array is assumed to be large enough. This isn't checked! Implemented in CContinuousActionGradientPolicy, CContinuousActionPolicyFromGradientFunction, CContinuousActionFeaturePolicy, CContinuousActionSigmoidPolicy, CContinuousActionQFunction, CCALinearFAQFunction, CContinuousTimeAndActionSigmoidVMGradientPolicy, CQFunction, CComposedQFunction, CTorchGradientFunction, CVFunctionFromGradientFunction, CQFunctionFromGradientFunction, and CFeatureVFunction. void CGradientUpdateFunction::updateGradient (  CFeatureList *  gradientFeatures, double  factor = 1.0 )    Does the preprocessing for the gradient update. When the weights are updated, the function updateGradient(CFeatureList *gradientFeatures, double factor = 1.0) is called. The gradient is at first saved in the local gradient buffer and then multiplied by the specified factor. If an eta calculator exists the eta calulator is applied to the gradient update, after that the user-interface function updateWeights is called. virtual void CGradientUpdateFunction::updateWeights (  CFeatureList *  dParams  )  [pure virtual]   Interface for updating the weights. Implemented in CContinuousActionGradientPolicy, CContinuousActionPolicyFromGradientFunction, CContinuousActionFeaturePolicy, CContinuousActionSigmoidPolicy, CCALinearFAQFunction, CContinuousTimeAndActionSigmoidVMGradientPolicy, CQFunction, CComposedQFunction, CTorchGradientFunction, CVFunctionFromGradientFunction, CQFunctionFromGradientFunction, and CFeatureVFunction. Member Data Documentation CAdaptiveEtaCalculator* CGradientUpdateFunction::etaCalc [protected] CFeatureList* CGradientUpdateFunction::localGradientFeatureBuffer [protected] The documentation for this class was generated from the following file: cgradientfunction.h