README

PHP的人工神经网络。具有使用梯度下降、动量、Sigmoid和双曲正切激活函数的在线反向传播学习功能。

关于

此库允许您构建和训练多层神经网络。首先定义网络结构。包括输入、输出、层数和隐藏节点数量。然后构建网络。使用邻接矩阵表示互连强度并初始化为小的随机值。然后将训练数据逐步呈现给网络。神经网络使用在线反向传播训练算法，该算法使用梯度下降下降误差曲线以调整互连强度。训练算法的目标是通过调整互连强度来减少全局误差。网络的全局误差通过均方误差计算。

您可以提供一个学习率参数和动量参数。学习率会影响神经网络收敛到最优解的速度。动量参数将帮助梯度下降避免在误差曲线上收敛到非最优解（称为局部最小值）。动量参数的正确大小将有助于找到全局最小值，但过大的值将阻止神经网络收敛到解。

训练好的神经网络可以保存到文件，并在以后激活时加载回来。

安装

$  composer require stephenlmoshea/neuralnetwork

示例

在具有两个输入和一个输出的三层神经网络上训练XOR函数

require __DIR__ . '/vendor/autoload.php';

use neuralnetwork\Network\FeedForward;
use neuralnetwork\Activation\Sigmoid;
use neuralnetwork\Train\Backpropagation;

//Create network with 2 input nodes, 2 hidden nodes, and 1 output node
//and set activation function
$network = new FeedForward([2, 2, 1], new Sigmoid());

//Define learning rate and momentum parameters for backpropagation algorithm
$ann = new Backpropagation($network, 0.7, 0.3);

//Provide XOR training data
$trainingSet = [
                    [0,0,0],
                    [0,1,1],
                    [1,0,1],
                    [1,1,0]
                ];

//Keep training the neural network until it converges
do {
    $ann->initialise();
    $result = $ann->train($trainingSet);
} while (!$result);

//Present [0,0] as network inputs and get the network output
$network->activate([0, 0]);
$outputs = $network->getOutputs();
echo $outputs[0]."\n";

//Present [0,1] as network inputs and get the network output
$network->activate([0, 1]);
$outputs = $network->getOutputs();
echo $outputs[0]."\n";

//Present [1,0] as network inputs and get the network output 
$network->activate([1, 0]);
$outputs = $network->getOutputs();
echo $outputs[0]."\n";

//Present [1,1] as network inputs and get the network output
$network->activate([1, 1]);
$outputs = $network->getOutputs();
echo $outputs[0]."\n";

使用双曲正切在三层神经网络上训练XOR函数

require __DIR__ . '/vendor/autoload.php';

use neuralnetwork\Network\FeedForward;
use neuralnetwork\Activation\HyperbolicTangent;
use neuralnetwork\Train\Backpropagation;

//Create network with 2 input nodes, 2 hidden nodes, and 1 output node
//and set activation function
$network = new FeedForward([2, 2, 1], new HyperbolicTangent());

//Define learning rate and momentum parameters for backpropagation algorithm
$ann = new Backpropagation($network, 0.7, 0.3, 0.001);

//Provide XOR training data
$trainingSet = [
                    [-1,-1,-1],
                    [-1,1,1],
                    [1,-1,1],
                    [1,1,-1]
                ];

//Keep training the neural network until it converges
do {
    $ann->initialise();
    $result = $ann->train($trainingSet);
} while (!$result);

//Present [-1,-1] as network inputs and get the network output
$network->activate([-1, -1]);
$outputs = $network->getOutputs();
echo $outputs[0]."\n";

//Present [-1,1] as network inputs and get the network output
$network->activate([-1, 1]);
$outputs = $network->getOutputs();
echo $outputs[0]."\n";

//Present [1,-1] as network inputs and get the network output 
$network->activate([1, -1]);
$outputs = $network->getOutputs();
echo $outputs[0]."\n";

//Present [1,1] as network inputs and get the network output
$network->activate([1, 1]);
$outputs = $network->getOutputs();
echo $outputs[0]."\n";

将训练好的神经网络保存到文件

require __DIR__ . '/vendor/autoload.php';

use neuralnetwork\Network\FeedForward;
use neuralnetwork\Activation\Sigmoid;
use neuralnetwork\Train\Backpropagation;

//Create network with 2 input nodes, 2 hidden nodes, and 1 output node
//and set activation function
$network = new FeedForward([2, 2, 1], new Sigmoid());

//Define learning rate and momentum parameters for backpropagation algorithm
$ann = new Backpropagation($network, 0.7, 0.3);

//Provide XOR training data
$trainingSet = [
                    [0,0,0],
                    [0,1,1],
                    [1,0,1],
                    [1,1,0]
                ];

//Keep training the neural network until it converges
do {
    $ann->initialise();
    $result = $ann->train($trainingSet);
} while (!$result);

$network->save('./network.txt');

从文件加载训练好的神经网络

require __DIR__ . '/vendor/autoload.php';

use neuralnetwork\Network\FeedForward;
use neuralnetwork\Activation\Sigmoid;
use neuralnetwork\Train\Backpropagation;

$network->load('./network.txt');

//Present [0,0] as network inputs and get the network output
$network->activate([0, 0]);
$outputs = $network->getOutputs();
echo $outputs[0]."\n";

//Present [0,1] as network inputs and get the network output
$network->activate([0, 1]);
$outputs = $network->getOutputs();
echo $outputs[0]."\n";

//Present [1,0] as network inputs and get the network output 
$network->activate([1, 0]);
$outputs = $network->getOutputs();
echo $outputs[0]."\n";

//Present [1,1] as network inputs and get the network output
$network->activate([1, 1]);
$outputs = $network->getOutputs();
echo $outputs[0]."\n";