Documentation Github NPM Package

Class

DQNAgent

DQNAgent(env, confignullable, randomSeednullable)

Implementation of DQN

Constructor

# new DQNAgent(env, confignullable, randomSeednullable)

Parameters:
Name Type Attributes Description
env Environment

The enviroment
config DQNAgentSettings <nullable>

The configuration
randomSeed number <nullable>

The configuration

View Source core/src/Agents/DQNAgent/DQNAgent.ts, line 14

Extends

Members

DQNNetwork

# network

Get the network

View Source core/src/Agents/DQNAgent/DQNAgent.ts, line 46

ReplayMemory

# replayMemory

Get the ReplayMemory

View Source core/src/Agents/DQNAgent/DQNAgent.ts, line 56

Methods

# abstract config() → {object|undefined}

get the configuration of the agent

Overrides:

View Source core/src/RLInterface/Agent.ts, line 68

object | undefined

# createNetwork() → {tf.Sequential}

Create a network

View Source core/src/Agents/DQNAgent/DQNAgent.ts, line 438

tf.Sequential

# decayEpsilon() → {void}

Decay the epsilon value

View Source core/src/Agents/DQNAgent/DQNAgent.ts, line 445

void

# abstract evalStep(state) → {string}

Method to select an action for prediction

Parameters:
Name Type Description
state object

The current game state
Overrides:

View Source core/src/RLInterface/Agent.ts, line 76

  • The action to select next
string

# abstract feed(prevState, takenAction, newState, payoff, contextInfo) → {Promise.<void>}

this method feed the new game state and reward back for the agent to update their algorithm

Parameters:
Name Type Description
prevState object

The previous game state
takenAction string

The action that was taken.
newState object

The new game state
payoff number

The gained payoff for the agent
contextInfo object

Through this object, additional information can be provided.
Overrides:

View Source core/src/RLInterface/Agent.ts, line 51

Promise.<void>

# abstract init() → {void}

initialize the agent

Overrides:

View Source core/src/RLInterface/Agent.ts, line 31

void

# abstract load(fileManager, optionsnullable) → {Promise.<void>}

load the Model

Parameters:
Name Type Attributes Description
fileManager FileStrategy

FileManager Strategy
options object <nullable>

the options zo use for loading
Overrides:

View Source core/src/RLInterface/PersistableAgent.ts, line 26

Promise.<void>

# abstract loadConfig(fileManager, optionsnullable) → {Promise.<void>}

load the config

Parameters:
Name Type Attributes Description
fileManager FileStrategy

FileManager Strategy
options object <nullable>

the options zo use for loading
Overrides:

View Source core/src/RLInterface/PersistableAgent.ts, line 35

Promise.<void>

# abstract log() → {void}

Interface method for loggin while training

Overrides:

View Source core/src/RLInterface/Agent.ts, line 83

void

# network() → {DQNNetwork}

Get the network

View Source core/src/Agents/DQNAgent/DQNAgent.ts, line 414

DQNNetwork

# replayMemory() → {ReplayMemory}

Get the ReplayMemory

View Source core/src/Agents/DQNAgent/DQNAgent.ts, line 422

ReplayMemory

# abstract reset() → {void}

Reset the agent

Overrides:

View Source core/src/RLInterface/Agent.ts, line 90

void

# abstract save(fileManager, optionsnullable) → {Promise.<void>}

Save the model

Parameters:
Name Type Attributes Description
fileManager FileStrategy

FileManager Strategy
options object <nullable>

the options to use for saving the agent
Overrides:

View Source core/src/RLInterface/PersistableAgent.ts, line 44

Promise.<void>

# abstract saveConfig(fileManager, optionsnullable) → {Promise.<void>}

Save the Config

Parameters:
Name Type Attributes Description
fileManager FileStrategy

FileManager Strategy
options object <nullable>

the options to use for saving the agent
Overrides:

View Source core/src/RLInterface/PersistableAgent.ts, line 53

Promise.<void>

# setConfig(confignullable, randomSeednullable) → {void}

Set The configuration of the agent after initailizing.

Parameters:
Name Type Attributes Description
config DQNAgentSettings <nullable>

The configuration.
randomSeed number <nullable>

The random seed.
Overrides:

View Source core/src/Agents/DQNAgent/DQNAgent.ts, line 431

void

# abstract step(state) → {string}

Method for selecting a new action for training

Parameters:
Name Type Description
state object

The current game state
Overrides:

View Source core/src/RLInterface/Agent.ts, line 39

  • The action to select next
string

# abstract trainingInitialized() → {boolean}

Get whether the agent was correctly initialized for training.

Overrides:

View Source core/src/RLInterface/Agent.ts, line 24

boolean