StandardDQN (deeplearning4j 1.0.0-beta7 API)

Skip navigation links

Prev Class
Next Class

All Classes

Summary:
Nested |
Field |
Constr |
Method

Detail:
Field |
Constr |
Method

java.lang.Object
- org.deeplearning4j.rl4j.learning.sync.qlearning.discrete.TDTargetAlgorithm.BaseTDTargetAlgorithm
- - org.deeplearning4j.rl4j.learning.sync.qlearning.discrete.TDTargetAlgorithm.BaseDQNAlgorithm
  - - org.deeplearning4j.rl4j.learning.sync.qlearning.discrete.TDTargetAlgorithm.StandardDQN

All Implemented Interfaces:

ITDTargetAlgorithm<Integer>
```
public class StandardDQN
extends BaseDQNAlgorithm
```
The Standard DQN algorithm based on "Playing Atari with Deep Reinforcement Learning" (https://arxiv.org/abs/1312.5602)

Author:

Alexandre Boulanger

Field Summary
- Fields inherited from class org.deeplearning4j.rl4j.learning.sync.qlearning.discrete.TDTargetAlgorithm.BaseDQNAlgorithm
  qNetworkNextObservation, targetQNetworkNextObservation
- Fields inherited from class org.deeplearning4j.rl4j.learning.sync.qlearning.discrete.TDTargetAlgorithm.BaseTDTargetAlgorithm
  gamma, qNetworkSource

Constructor Summary

Constructors
Constructor and Description
`StandardDQN(TargetQNetworkSource qTargetNetworkSource, double gamma)`
`StandardDQN(TargetQNetworkSource qTargetNetworkSource, double gamma, double errorClamp)`

Method Summary

All Methods Instance Methods Concrete Methods
Modifier and Type	Method and Description
`protected double`	`computeTarget(int batchIdx, double reward, boolean isTerminal)` In litterature, this corresponds to: Q(s_t, a_t) = R_{t+1} + \gamma * max_{a}Q_{tar}(s_{t+1}, a)
`protected void`	`initComputation(INDArray observations, INDArray nextObservations)` Called just before the calculation starts

Methods inherited from class org.deeplearning4j.rl4j.learning.sync.qlearning.discrete.TDTargetAlgorithm.BaseTDTargetAlgorithm
computeTDTargets

Methods inherited from class java.lang.Object
clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

- Constructor Detail
  - StandardDQN
```
public StandardDQN(TargetQNetworkSource qTargetNetworkSource,
                   double gamma)
```
  - StandardDQN
```
public StandardDQN(TargetQNetworkSource qTargetNetworkSource,
                   double gamma,
                   double errorClamp)
```
- Method Detail
  - initComputation
```
protected void initComputation(INDArray observations,
                               INDArray nextObservations)
```
    Description copied from class: BaseTDTargetAlgorithm
    
    Called just before the calculation starts
    
    Overrides:
    
    initComputation in class BaseDQNAlgorithm
    
    Parameters:
    
    observations - A INDArray of all observations stacked on dimension 0
    
    nextObservations - A INDArray of all next observations stacked on dimension 0
  - computeTarget
```
protected double computeTarget(int batchIdx,
                               double reward,
                               boolean isTerminal)
```
    In litterature, this corresponds to:
    Q(s_t, a_t) = R_{t+1} + \gamma * max_{a}Q_{tar}(s_{t+1}, a)
    
    Specified by:
    
    computeTarget in class BaseTDTargetAlgorithm
    
    Parameters:
    
    batchIdx - The index in the batch of the current transition
    
    reward - The reward of the current transition
    
    isTerminal - True if it's the last transition of the "game"
    
    Returns:
    
    The estimated Q-Value

Skip navigation links

Prev Class
Next Class

All Classes

Summary:
Nested |
Field |
Constr |
Method

Detail:
Field |
Constr |
Method

Copyright © 2020. All rights reserved.