dqagent.md

ReImproveJS

Globals / DQAgent /

Class: DQAgent

Hierarchy

• AbstractAgent

  • DQAgent

Index

Constructors

• constructor

Properties

• actionsBuffer
• agentConfig
• currentReward
• done
• forwardPasses
• inputsBuffer
• lossesHistory
• memory
• model
• netInputWindowSize
• rewardsHistory
• statesBuffer

Accessors

• AgentConfig
• Name

Methods

• addReward
• createNeuralNetInput
• createTrainingDataFromMemento
• getTrackingInformation
• infer
• learn
• listen
• memorize
• policy
• reset
• setAgentConfig
• setReward

Constructors

constructor

+ new DQAgent(model: Model, agentConfig?: DQAgentConfig, name?: string): DQAgent

Overrides AbstractAgent.constructor

Defined in reimprove/algorithms/deepq/dqagent.ts:33

Parameters:

Name	Type
model	Model
agentConfig?	DQAgentConfig
name?	string

Returns: DQAgent

---

Properties

Private actionsBuffer

● actionsBuffer: Array<number>

Defined in reimprove/algorithms/deepq/dqagent.ts:23

---

Protected agentConfig

● agentConfig: AgentConfig

Inherited from AbstractAgent.agentConfig

Defined in reimprove/algorithms/abstract_agent.ts:10

---

Private currentReward

● currentReward: number

Defined in reimprove/algorithms/deepq/dqagent.ts:21

---

Private done

● done: boolean

Defined in reimprove/algorithms/deepq/dqagent.ts:20

---

Private forwardPasses

● forwardPasses: number

Defined in reimprove/algorithms/deepq/dqagent.ts:33

---

Private inputsBuffer

● inputsBuffer: Array<MementoTensor>

Defined in reimprove/algorithms/deepq/dqagent.ts:25

---

Private lossesHistory

● lossesHistory: TypedWindow‹number›

Defined in reimprove/algorithms/deepq/dqagent.ts:27

---

Private memory

● memory: Memory

Defined in reimprove/algorithms/deepq/dqagent.ts:31

---

Private model

● model: Model

Defined in reimprove/algorithms/deepq/dqagent.ts:36

---

Private netInputWindowSize

● netInputWindowSize: number

Defined in reimprove/algorithms/deepq/dqagent.ts:29

---

Private rewardsHistory

● rewardsHistory: TypedWindow‹number›

Defined in reimprove/algorithms/deepq/dqagent.ts:28

---

Private statesBuffer

● statesBuffer: Array<Tensor>

Defined in reimprove/algorithms/deepq/dqagent.ts:24

---

Accessors

AgentConfig

● get AgentConfig(): DQAgentConfig

Overrides AbstractAgent.AgentConfig

Defined in reimprove/algorithms/deepq/dqagent.ts:204

Returns: DQAgentConfig

● set AgentConfig(config: DQAgentConfig): void

Overrides AbstractAgent.AgentConfig

Defined in reimprove/algorithms/deepq/dqagent.ts:208

Parameters:

Name	Type
config	DQAgentConfig

Returns: void

---

Name

● get Name(): string

Inherited from AbstractAgent.Name

Defined in reimprove/algorithms/abstract_agent.ts:19

Returns: string

● set Name(name: string): void

Inherited from AbstractAgent.Name

Defined in reimprove/algorithms/abstract_agent.ts:20

Parameters:

Name	Type
name	string

Returns: void

---

Methods

addReward

▸ addReward(value: number): void

Defined in reimprove/algorithms/deepq/dqagent.ts:189

Parameters:

Name	Type
value	number

Returns: void

---

Private createNeuralNetInput

▸ createNeuralNetInput(input: Tensor): Tensor

Defined in reimprove/algorithms/deepq/dqagent.ts:55

Parameters:

Name	Type
input	Tensor

Returns: Tensor

---

createTrainingDataFromMemento

▸ createTrainingDataFromMemento(memento: Memento, gamma: number, alpha: number): object

Defined in reimprove/algorithms/deepq/dqagent.ts:142

Parameters:

Name	Type
memento	Memento
gamma	number
alpha	number

Returns: object

• x: Tensor

• y: Tensor

---

getTrackingInformation

▸ getTrackingInformation(): AgentTrackingInformation

Overrides AbstractAgent.getTrackingInformation

Defined in reimprove/algorithms/deepq/dqagent.ts:212

Returns: AgentTrackingInformation

---

infer

▸ infer(input: number[] | number[][], epsilon: number, keepTensors: boolean): number

Overrides AbstractAgent.infer

Defined in reimprove/algorithms/deepq/dqagent.ts:80

Parameters:

Name	Type	Default
input	number[] | number[][]	-
epsilon	number	-
keepTensors	boolean	true

Returns: number

---

learn

▸ learn(gamma: number, alpha: number): Promise<void>

Defined in reimprove/algorithms/deepq/dqagent.ts:162

Parameters:

Name	Type
gamma	number
alpha	number

Returns: Promise<void>

---

listen

▸ listen(input: number[] | number[][], epsilon: number): number

Defined in reimprove/algorithms/deepq/dqagent.ts:155

Parameters:

Name	Type
input	number[] | number[][]
epsilon	number

Returns: number

---

memorize

▸ memorize(): void

Defined in reimprove/algorithms/deepq/dqagent.ts:127

Returns: void

---

Private policy

▸ policy(input: Tensor): number

Defined in reimprove/algorithms/deepq/dqagent.ts:76

Parameters:

Name	Type
input	Tensor

Returns: number

---

reset

▸ reset(): void

Overrides AbstractAgent.reset

Defined in reimprove/algorithms/deepq/dqagent.ts:197

Returns: void

---

Protected setAgentConfig

▸ setAgentConfig(config: AgentConfig): void

Inherited from AbstractAgent.setAgentConfig

Defined in reimprove/algorithms/abstract_agent.ts:17

Parameters:

Name	Type
config	AgentConfig

Returns: void

---

setReward

▸ setReward(value: number): void

Defined in reimprove/algorithms/deepq/dqagent.ts:193

Parameters:

Name	Type
value	number

Returns: void

---