AsyncLearning< WorkerType, EnvironmentType, NetworkType, UpdaterType, PolicyType > Class Template Reference

Wrapper of various asynchronous learning algorithms, e.g. More...

Public Member Functions
	AsyncLearning (TrainingConfig config, NetworkType network, PolicyType policy, UpdaterType updater=UpdaterType(), EnvironmentType environment=EnvironmentType())
	Construct an instance of the given async learning algorithm. More...
TrainingConfig &	Config ()
	Get training config. More...
const TrainingConfig &	Config () const
	Modify training config. More...
EnvironmentType &	Environment ()
	Get the environment. More...
const EnvironmentType &	Environment () const
	Modify the environment. More...
NetworkType &	Network ()
	Get learning network. More...
const NetworkType &	Network () const
	Modify learning network. More...
PolicyType &	Policy ()
	Get behavior policy. More...
const PolicyType &	Policy () const
	Modify behavior policy. More...
template < typename Measure >
void	Train (Measure &measure)
	Starting async training. More...
UpdaterType &	Updater ()
	Get optimizer. More...
const UpdaterType &	Updater () const
	Modify optimizer. More...

Detailed Description

template

<

typename

WorkerType

,

typename

EnvironmentType

,

typename

NetworkType

,

typename

UpdaterType

,

typename

PolicyType

>

class mlpack::rl::AsyncLearning< WorkerType, EnvironmentType, NetworkType, UpdaterType, PolicyType >

Wrapper of various asynchronous learning algorithms, e.g.

async one-step Q-learning, async one-step Sarsa, async n-step Q-learning and async advantage actor-critic.

For more details, see the following:

@inproceedings{mnih2016asynchronous,
  title     = {Asynchronous methods for deep reinforcement learning},
  author    = {Mnih, Volodymyr and Badia, Adria Puigdomenech and Mirza,
               Mehdi and Graves, Alex and Lillicrap, Timothy and Harley,
               Tim and Silver, David and Kavukcuoglu, Koray},
  booktitle = {International Conference on Machine Learning},
  pages     = {1928--1937},
  year      = {2016}
}

Template Parameters

WorkerType	The type of the worker.
EnvironmentType	The type of reinforcement learning task.
NetworkType	The type of the network model.
UpdaterType	The type of the optimizer.
PolicyType	The type of the behavior policy.

Definition at line 57 of file async_learning.hpp.

Constructor & Destructor Documentation

AsyncLearning()

AsyncLearning	(	TrainingConfig	config,
		NetworkType	network,
		PolicyType	policy,
		UpdaterType	updater = UpdaterType(),
		EnvironmentType	environment = EnvironmentType()
	)

Construct an instance of the given async learning algorithm.

Parameters

config	Hyper-parameters for training.
network	The network model.
policy	The behavior policy.
updater	The optimizer.
environment	The reinforcement learning task.

Member Function Documentation

Config() [1/2]

TrainingConfig& Config ( )

inline

Get training config.

Definition at line 92 of file async_learning.hpp.

Config() [2/2]

const TrainingConfig& Config ( ) const

inline

Modify training config.

Definition at line 94 of file async_learning.hpp.

Environment() [1/2]

EnvironmentType& Environment ( )

inline

Get the environment.

Definition at line 112 of file async_learning.hpp.

Environment() [2/2]

const EnvironmentType& Environment ( ) const

inline

Modify the environment.

Definition at line 114 of file async_learning.hpp.

Network() [1/2]

NetworkType& Network ( )

inline

Get learning network.

Definition at line 97 of file async_learning.hpp.

Network() [2/2]

const NetworkType& Network ( ) const

inline

Modify learning network.

Definition at line 99 of file async_learning.hpp.

Policy() [1/2]

PolicyType& Policy ( )

inline

Get behavior policy.

Definition at line 102 of file async_learning.hpp.

Policy() [2/2]

const PolicyType& Policy ( ) const

inline

Modify behavior policy.

Definition at line 104 of file async_learning.hpp.

Train()

void Train

(

Measure &

measure

)

Starting async training.

Template Parameters

Measure

The type of the measurement. It should be a callable object like bool foo(double reward); where reward is the total reward of a deterministic test episode, and the return value should indicate whether the training process is completed.

Parameters

measure

The measurement instance.

Updater() [1/2]

UpdaterType& Updater ( )

inline

Get optimizer.

Definition at line 107 of file async_learning.hpp.

Updater() [2/2]

const UpdaterType& Updater ( ) const

inline

Modify optimizer.

Definition at line 109 of file async_learning.hpp.

---

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

• /home/ryan/src/mlpack.org/_src/mlpack-git/src/mlpack/methods/reinforcement_learning/async_learning.hpp