/**
* Construct an ADALINE trainer.
*
* @param BasicNetwork network
* The network to train.
* @param MLDataSet training
* The training data.
* @param double learningRate
* The learning rate.
*/
public function __construct(BasicNetwork $network, MLDataSet $training, $learningRate)
{
parent::__construct(TrainingImplementationType::Iterative);
if ($network->getLayerCount() > 2) {
throw new NeuralNetworkError("An ADALINE network only has two layers.");
}
$this->network = $network;
$this->training = $training;
$this->learningRate = $learningRate;
}
/**
* Generate the network.
*
* @return MLMethod The generated network.
*/
public function generate()
{
$network = new BasicNetwork();
$inputLayer = new BasicLayer(new ActivationLinear(), true, $this->inputNeurons);
$outputLayer = new BasicLayer(new ActivationLinear(), false, $this->outputNeurons);
$network->addLayer($inputLayer);
$network->addLayer($outputLayer);
$network->getStructure()->finalizeStructure();
(new RangeRandomizer(-0.5, 0.5))->randomizeMLMethod($network);
return $network;
}
/**
* Randomize one level of a neural network.
*
* @param
* BasicNetwork network
* The network to randomize
* @param
* int fromLayer
* The from level to randomize.
*/
public function randomizeNetwork(BasicNetwork $network, $fromLayer)
{
$fromCount = $network->getLayerTotalNeuronCount($fromLayer);
$toCount = $network->getLayerNeuronCount($fromLayer + 1);
for ($fromNeuron = 0; $fromNeuron < $fromCount; ++$fromNeuron) {
for ($toNeuron = 0; $toNeuron < $toCount; ++$toNeuron) {
$v = $network->getWeight($fromLayer, $fromNeuron, $toNeuron);
$v = $this->randomizeDouble($v);
$network->setWeight($fromLayer, $fromNeuron, $toNeuron, $v);
}
}
}
