Neuron.java

public class Neuron {

	// main constructor
	public Neuron(int prev_n_neurons, java.util.Random rand)
	{
		// each neuron know the weights of each connection 
		// with neurons of the previous layer
		_synapticWeights = new float[prev_n_neurons];

		// set default weights
		for (int i = 0; i < prev_n_neurons; ++i)
			_synapticWeights[i] = rand.nextFloat() - 0.5f;
	}

	// activate the neuron with given inputs, return the output
	public float activate(float inputs[])
	{
		_activation = 0.0f;
		assert(inputs.length == _synapticWeights.length);

		for (int i = 0; i < inputs.length; ++i) // dot product (produit scalaire)
			_activation += inputs[i] * _synapticWeights[i];

		// phi(_activation), our activation function (tanh(x))
		return 2.0f / (1.0f + (float) Math.exp((-_activation) * lambda)) - 1.0f;
	}

	public float getActivationDerivative() // dphi(_activation)
	{
		float expmlx = (float) Math.exp(lambda * _activation);
		return 2 * lambda * expmlx / ((1 + expmlx) * (1 + expmlx));
	}

	public float[] getSynapticWeights() { return _synapticWeights; }
	public float getSynapticWeight(int i) { return _synapticWeights[i]; }
	public void setSynapticWeight(int i, float v) { _synapticWeights[i] = v; }

	// --------
	private float _activation;
	private float[] _synapticWeights;

	// parameter of the sigmoid
	static final float lambda = 1.5f;
}