mlpack::rl::CategoricalDQN< OutputLayerType, InitType, NetworkType > Class Template Reference

Implementation of the Categorical Deep Q-Learning network. More...

#include <categorical_dqn.hpp>

Public Member Functions
	CategoricalDQN ()
	Default constructor.
	CategoricalDQN (const int inputDim, const int h1, const int h2, const int outputDim, TrainingConfig config, const bool isNoisy=false, InitType init=InitType(), OutputLayerType outputLayer=OutputLayerType())
	Construct an instance of CategoricalDQN class. More...
	CategoricalDQN (NetworkType &network, TrainingConfig config, const bool isNoisy=false)
	Construct an instance of CategoricalDQN class from a pre-constructed network. More...
void	Predict (const arma::mat state, arma::mat &actionValue)
	Predict the responses to a given set of predictors. More...
void	Forward (const arma::mat state, arma::mat &dist)
	Perform the forward pass of the states in real batch mode. More...
void	ResetParameters ()
	Resets the parameters of the network.
void	ResetNoise ()
	Resets noise of the network, if the network is of type noisy.
const arma::mat &	Parameters () const
	Return the Parameters.
arma::mat &	Parameters ()
	Modify the Parameters.
void	Backward (const arma::mat state, arma::mat &lossGradients, arma::mat &gradient)
	Perform the backward pass of the state in real batch mode. More...

Detailed Description

template<typename OutputLayerType = EmptyLoss<>, typename InitType = GaussianInitialization, typename NetworkType = FFN<OutputLayerType, InitType>>
class mlpack::rl::CategoricalDQN< OutputLayerType, InitType, NetworkType >

Implementation of the Categorical Deep Q-Learning network.

For more information, see the following.

@misc{bellemare2017distributional,
  author  = {Marc G. Bellemare, Will Dabney, Rémi Munos},
  title   = {A Distributional Perspective on Reinforcement Learning},
  year    = {2017},
  url     = {http://arxiv.org/abs/1707.06887}
}

Template Parameters

OutputLayerType	The output layer type of the network.
InitType	The initialization type used for the network.
NetworkType	The type of network used for simple dqn.

Constructor & Destructor Documentation

CategoricalDQN() [1/2]

template<typename OutputLayerType = EmptyLoss<>, typename InitType = GaussianInitialization, typename NetworkType = FFN<OutputLayerType, InitType>>

mlpack::rl::CategoricalDQN< OutputLayerType, InitType, NetworkType >::CategoricalDQN ( const int  inputDim, const int  h1, const int  h2, const int  outputDim, TrainingConfig  config, const bool  isNoisy = false, InitType  init = InitType(), OutputLayerType  outputLayer = OutputLayerType()  )

inline

Construct an instance of CategoricalDQN class.

Parameters

inputDim	Number of inputs.
h1	Number of neurons in hiddenlayer-1.
h2	Number of neurons in hiddenlayer-2.
outputDim	Number of neurons in output layer.
config	Hyper-parameters for categorical dqn.
isNoisy	Specifies whether the network needs to be of type noisy.
init	Specifies the initialization rule for the network.
outputLayer	Specifies the output layer type for network.

CategoricalDQN() [2/2]

template<typename OutputLayerType = EmptyLoss<>, typename InitType = GaussianInitialization, typename NetworkType = FFN<OutputLayerType, InitType>>

mlpack::rl::CategoricalDQN< OutputLayerType, InitType, NetworkType >::CategoricalDQN ( NetworkType &  network, TrainingConfig  config, const bool  isNoisy = false  )

inline

Construct an instance of CategoricalDQN class from a pre-constructed network.

Parameters

network	The network to be used by CategoricalDQN class.
config	Hyper-parameters for categorical dqn.
isNoisy	Specifies whether the network needs to be of type noisy.

Member Function Documentation

Backward()

template<typename OutputLayerType = EmptyLoss<>, typename InitType = GaussianInitialization, typename NetworkType = FFN<OutputLayerType, InitType>>

void mlpack::rl::CategoricalDQN< OutputLayerType, InitType, NetworkType >::Backward ( const arma::mat  state, arma::mat &  lossGradients, arma::mat &  gradient  )

inline

Perform the backward pass of the state in real batch mode.

Parameters

state	The input state.
lossGradients	The loss gradients.
gradient	The gradient.

Forward()

template<typename OutputLayerType = EmptyLoss<>, typename InitType = GaussianInitialization, typename NetworkType = FFN<OutputLayerType, InitType>>

void mlpack::rl::CategoricalDQN< OutputLayerType, InitType, NetworkType >::Forward ( const arma::mat  state, arma::mat &  dist  )

inline

Perform the forward pass of the states in real batch mode.

Parameters

state	The input state.
dist	The predicted distributions.

Predict()

template<typename OutputLayerType = EmptyLoss<>, typename InitType = GaussianInitialization, typename NetworkType = FFN<OutputLayerType, InitType>>

void mlpack::rl::CategoricalDQN< OutputLayerType, InitType, NetworkType >::Predict ( const arma::mat  state, arma::mat &  actionValue  )

inline

Predict the responses to a given set of predictors.

The responses will reflect the output of the given output layer as returned by the output layer function.

If you want to pass in a parameter and discard the original parameter object, be sure to use std::move to avoid unnecessary copy.

Parameters

state	Input state.
actionValue	Matrix to put output action values of states input.

---

The documentation for this class was generated from the following file:

• src/mlpack/methods/reinforcement_learning/q_networks/categorical_dqn.hpp