NeuralNet (deeplearning4j 1.0.0-beta7 API)

All Known Subinterfaces:

IActorCritic<NN>, IDQN<NN>

All Known Implementing Classes:

ActorCriticCompGraph, ActorCriticSeparate, DQN
```
public interface NeuralNet<NN extends NeuralNet>
```
Author:

rubenfiszel (ruben.fiszel@epfl.ch) on 8/5/16. Factorisation between ActorCritic and DQN neural net. Useful for AsyncLearning and Thread code.

Method Summary

All Methods Instance Methods Abstract Methods
Modifier and Type	Method and Description
`void`	`applyGradient(Gradient[] gradients, int batchSize)` update the params from the gradients and the batchSize
`NN`	`clone()` clone the Neural Net with the same paramaeters
`void`	`copy(NN from)` copy the parameters from a neural net
`void`	`fit(INDArray input, INDArray[] labels)` fit from input and labels
`double`	`getLatestScore()` latest score from lastest fit
`NeuralNetwork[]`	`getNeuralNetworks()` Returns the underlying MultiLayerNetwork or ComputationGraph objects.
`Gradient[]`	`gradient(INDArray input, INDArray[] labels)` Calculate the gradients from input and label (target) of all outputs
`boolean`	`isRecurrent()` returns true if this is a recurrent network
`INDArray[]`	`outputAll(INDArray batch)`
`void`	`reset()` required for recurrent networks during init
`void`	`save(OutputStream os)` save the neural net into an OutputStream
`void`	`save(String filename)` save the neural net into a filename

- Method Detail
  - getNeuralNetworks
```
NeuralNetwork[] getNeuralNetworks()
```
    Returns the underlying MultiLayerNetwork or ComputationGraph objects.
  - isRecurrent
```
boolean isRecurrent()
```
    returns true if this is a recurrent network
  - reset
```
void reset()
```
    required for recurrent networks during init
  - outputAll
```
INDArray[] outputAll(INDArray batch)
```
    Parameters:
    
    batch - batch to evaluate
    
    Returns:
    
    evaluation by the model of the input by all outputs
  - clone
```
NN clone()
```
    clone the Neural Net with the same paramaeters
    
    Returns:
    
    the cloned neural net
  - copy
```
void copy(NN from)
```
    copy the parameters from a neural net
    
    Parameters:
    
    from - where to copy parameters
  - gradient
```
Gradient[] gradient(INDArray input,
                    INDArray[] labels)
```
    Calculate the gradients from input and label (target) of all outputs
    
    Parameters:
    
    input - input batch
    
    labels - target batch
    
    Returns:
    
    the gradients
  - fit
```
void fit(INDArray input,
         INDArray[] labels)
```
    fit from input and labels
    
    Parameters:
    
    input - input batch
    
    labels - target batch
  - applyGradient
```
void applyGradient(Gradient[] gradients,
                   int batchSize)
```
    update the params from the gradients and the batchSize
    
    Parameters:
    
    gradients - gradients to apply the gradient from
    
    batchSize - batchSize from which the gradient was calculated on (similar to nstep)
  - getLatestScore
```
double getLatestScore()
```
    latest score from lastest fit
    
    Returns:
    
    latest score
  - save
```
void save(OutputStream os)
   throws IOException
```
    save the neural net into an OutputStream
    
    Parameters:
    
    os - OutputStream to save in
    
    Throws:
    
    IOException
  - save
```
void save(String filename)
   throws IOException
```
    save the neural net into a filename
    
    Parameters:
    
    filename - filename to save in
    
    Throws:
    
    IOException

Interface NeuralNet<NN extends NeuralNet>

Method Summary

Method Detail

getNeuralNetworks

isRecurrent

reset

outputAll

clone

copy

gradient

fit

applyGradient

getLatestScore

save

save