public void Learn(PerceptronLayer layer, float learningRate = 0.7f)
{
for (var i = 0; i < ps.Count; ++i)
{
var p = ps [i];
var err = 0f;
for (var j = 0; j < layer.ps.Count; ++j)
{
var p2 = layer.ps [j];
err += p2.W [i] * p2.Error;
}
var target = p.Output + err;
/*
* if (target > 1) {
* target = 1;
* }
* if (target < -1) {
* target = -1;
* }
*/
p.Learn(target, learningRate);
//p.LearnWithError (err, learningRate);
}
}
void Start()
{
var layer = new PerceptronLayer(2);
layer.Add(new Perceptron(2));
layer.Add(new Perceptron(2));
//layer.Add (new Perceptron (2));
var layer2 = new PerceptronLayer(2);
layer2.Add(new Perceptron(2));
/*
* layer2.Add (new Perceptron (3));
*
* var layer3 = new PerceptronLayer (2);
* layer3.Add (new Perceptron (2));
*/
// 總共3神經元就能訓練xor
p.Add(layer);
p.Add(layer2);
// p.Add (layer3);
for (var i = 0; i < pixels.GetLength(0); ++i)
{
for (var j = 0; j < pixels.GetLength(1); ++j)
{
pixels [i, j] = Instantiate(pixel, new Vector3(i, j, 0), new Quaternion()) as GameObject;
}
}
}
void Start()
{
float[] state = TeacherCar.State;
float[] action = TeacherCar.Action;
bpn = new MultiLayerPerceptron(state.Length);
var layer = new PerceptronLayer(state.Length);
layer.Add(new Perceptron(state.Length));
layer.Add(new Perceptron(state.Length));
layer.Add(new Perceptron(state.Length));
var layer2 = new PerceptronLayer(3);
layer2.Add(new Perceptron(3));
layer2.Add(new Perceptron(3));
var layer3 = new PerceptronLayer(2);
for (var i = 0; i < action.Length; ++i)
{
layer3.Add(new Perceptron(2));
}
bpn.Add(layer);
bpn.Add(layer2);
bpn.Add(layer3);
/*
* for (var i = 0; i < 1000; ++i) {
* bpn.Input = new float[]{ 0, 0, 0 };
* bpn.Feed ();
* bpn.Learn (new float[]{ 1, 1 });
*
* bpn.Input = new float[]{ 1, 1, 1 };
* bpn.Feed ();
* bpn.Learn (new float[]{ 0, 0 });
*
* bpn.Input = new float[]{ 1, 0, 1 };
* bpn.Feed ();
* bpn.Learn (new float[]{ 0.5f, 0.5f });
* }
*
* bpn.Input = new float[]{ 0, 0, 0 };
* bpn.Feed ();
* var currAction = bpn.Output;
* print (currAction[0]+","+currAction[1]);
*
* bpn.Input = new float[]{ 1, 1, 1 };
* bpn.Feed ();
* currAction = bpn.Output;
* print (currAction[0]+","+currAction[1]);
*
* bpn.Input = new float[]{ 1, 0, 1 };
* bpn.Feed ();
* currAction = bpn.Output;
* print (currAction[0]+","+currAction[1]);
*/
}
public void Learn(float[] targets, float learningRate = 0.7f)
{
PerceptronLayer outerLayer = null;
for (var i = layers.Count - 1; i >= 0; --i)
{
var layer = layers [i];
if (i == layers.Count - 1)
{
layer.Learn(targets, learningRate);
}
else
{
layer.Learn(outerLayer, learningRate);
}
outerLayer = layer;
}
}
public void Add(PerceptronLayer layer)
{
layers.Add(layer);
}
