/// <summary>
/// ns 记录从输入层开始，每层的节点个数。
/// withBias 为真，表示所有的节点均带有偏置。
/// fp 激活函数及其导数
/// u 训练步长
/// ranndomOrNull 初始化权值的方法
/// </summary>
public Network(List <int> ns, bool withBias, Function_Pair fp, double u, bool RandomOrNull)
{
Layers = new List <Layer>();
this.withBias = withBias;
for (int i = 0; i < ns.Count; i++)
{
int n = ns[i];
bool IsInput = false;
bool IsOutput = false;
if (i == 0)
{
IsInput = true;
}
if (i == ns.Count - 1)
{
IsOutput = true;
withBias = false;//输出层不会有偏置。
}
Layer layer = new Layer(withBias, n, IsInput, IsOutput, this, i, fp);
Layers.Add(layer);
}
this.u = u;
GodFunctions = new List <GodFunction>();
InitializeWeight(RandomOrNull);
}
public Node(bool IsBias, Layer layer, Function_Pair fp)
{
this.IsBias = IsBias;
this.layer = layer;
if (!this.IsInput)
{
this.FP = fp; // 输入层无需激活函数。
Inputs = new List <Node>();
weights = new List <double>();
}
}
/// <summary>
/// 是否带偏置，该层的节点个数（不包含偏置在内）
/// 是否为输入层，是否为输出层。
/// 网络，该层的层数。激活函数。
/// </summary>
public Layer(bool withBias, int N_nodes, bool IsInput, bool IsOutput, Network net, int Nth_layer, Function_Pair fp)
{
nodes = new List <Node>();
this.IsInput = IsInput;
this.IsOutput = IsOutput;
this.net = net;
this.Nth_layer = Nth_layer;
if (withBias)
{
nodes.Add(new Node(true, this, fp));
}
for (int i = 0; i < N_nodes; i++)
{
Node node = new Node(false, this, fp);
nodes.Add(node);
if (!node.IsInput)
{
foreach (Node pnode in this.PreviousLayer.nodes)
{
node.Inputs.Add(pnode);
node.weights.Add(new double());
}
}
}
}
