com.github.mrdimosthenis.synapses

Class Net

java.lang.Object

com.github.mrdimosthenis.synapses.Net

public class Net
extends java.lang.Object

The constructors and methods of neural networks.

Create a neural network:

 
 Net net = new Net(new int[]{3, 4, 5});
 
 

Get the prediction for an input:

 
 net.predict(new double[]{0.4, 0.05, 0.2});
 
 

Fit network to a single observation:

 
 net.fit(0.1, new double[]{0.4, 0.05, 0.2}, new double[]{0.03. 0.8});
 
 

Get the JSON representation of the network:

 
 net.json();
 
 

Constructor Summary

Constructors

Constructor and Description
Net(int[] layerSizes) Creates a random neural network.
Net(int[] layerSizes, java.util.function.IntFunction<Fun> activationF, java.util.function.IntFunction<java.lang.Double> weightInitF) Creates a neural network.
Net(int[] layerSizes, java.lang.Long seed) Creates a non-random neural network.
Net(java.lang.String json) Cretaes a neural network by parsing its JSON representation.

Method Summary

Modifier and Type	Method and Description
double[]	errors(double[] inputValues, double[] expectedOutput, boolean inParallel)
void	fit(double learningRate, double[] inputValues, double[] expectedOutput) Adjust the weights of the neural network to the provided observation.
void	fitPar(double learningRate, double[] inputValues, double[] expectedOutput) Adjust the weights of the neural network to the provided observation.
java.lang.String	json() The JSON representation of the neural network.
double[]	parPredict(double[] inputValues) Makes a prediction for the provided input.
double[]	predict(double[] inputValues) Makes a prediction for the provided input.
java.lang.String	svg() An SVG representation of the neural network.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

Net

public Net(int[] layerSizes,
           java.util.function.IntFunction<Fun> activationF,
           java.util.function.IntFunction<java.lang.Double> weightInitF)

Creates a neural network.

Parameters:

layerSizes - The size of each layer.

activationF - A function that accepts the index of a layer and returns an activation function for its neurons.

weightInitF - A function that accepts the index of a layer and returns a weight for the synapses of its neurons.

Net

public Net(int[] layerSizes)

Creates a random neural network.

The activation function of all neurons is sigmoid. The weight distribution of the synapses is normal between -1.0 and 1.0.

Parameters:

layerSizes - The size of each layer. The first number in the list defines the size of the input layer. The last number in the list defines the size of the output layer. In order for a neural network to be deep, the list should contain more than two numbers.

Net

public Net(int[] layerSizes,
           java.lang.Long seed)

Creates a non-random neural network.

Calling this function with the same parameters multiple times, should always return the same neural network. The activation function of the nodes is sigmoid. The weight distribution of the synapses is normal between -1.0 and 1.0.

Parameters:

layerSizes - The size of each layer.

seed - A number used to initialize the internal pseudorandom number generator.

Net

public Net(java.lang.String json)

Cretaes a neural network by parsing its JSON representation.

Parameters:

json - The JSON representation of a neural network.

Method Detail

predict

public double[] predict(double[] inputValues)

Makes a prediction for the provided input.

Parameters:

inputValues - The values of the features. Their size should be equal to the size of the input layer.

Returns:

The prediction. It's size should be equal to the size of the output layer.

parPredict

public double[] parPredict(double[] inputValues)

Makes a prediction for the provided input.

The calculation is performed in parallel. When the neural network has huge layers, the parallel calculation boosts the performance.

Parameters:

inputValues - The values of the features. Their size should be equal to the size of the input layer.

Returns:

The prediction. It's size should be equal to the size of the output layer.

errors

public double[] errors(double[] inputValues,
                       double[] expectedOutput,
                       boolean inParallel)

fit

public void fit(double learningRate,
                double[] inputValues,
                double[] expectedOutput)

Adjust the weights of the neural network to the provided observation.

In order for it to be trained, it should fit with multiple observations.

Parameters:

learningRate - A number that controls how much the weights are adjusted to the observation.

inputValues - The feature values of the observation.

expectedOutput - The expected output of the observation. It's size should be equal to the size of the output layer.

fitPar

public void fitPar(double learningRate,
                   double[] inputValues,
                   double[] expectedOutput)

Adjust the weights of the neural network to the provided observation.

The calculation is performed in parallel. When the neural network has huge layers, the parallel calculation boosts the performance.

Parameters:

learningRate - A number that controls how much the weights are adjusted to the observation.

inputValues - The feature values of the observation.

expectedOutput - The expected output of the observation. It's size should be equal to the size of the output layer.

json

public java.lang.String json()

The JSON representation of the neural network.

Returns:

The JSON representation of the neural network.

svg

public java.lang.String svg()

An SVG representation of the neural network.

Returns:

The SVG representation of the neural network. The color of each neuron depends on its activation function while the transparency of the synapses depends on their weight.