private void LoadNextBatchSet(int offset)
{
if (offset < layers[0].matrix.GetLength(0))
{
X = layers[0].matrix.GetBatch(_batchSize, offset);
X = X.AddFeatureBias(1);
Y = AllY.GetBatch(_batchSize, offset);
}
}
public double[,] CheckAnswer(double[,] TX)
{
double[][,] At = new double[layers.Length - 1][, ];
double[][,] Zt = new double[layers.Length - 1][, ];
double[,] TestedX = TX.AddFeatureBias(1);
for (int i = 1; i < layers.Length; i++)
{
if (i == 1)
{
At[i - 1] = new double[TestedX.GetLength(0), layers[i].neurons];
}
else if (i == layers.Length - 1)
{
At[i - 1] = new double[layers[i - 1].neurons, layers[i].matrix.GetLength(1)];
}
else
{
At[i - 1] = new double[layers[i - 1].neurons, layers[i].neurons];
}
}
for (int i = 0; i < layers.Length - 1; i++)
{
if (i == 0)
{
Zt[i] = TestedX.Mul(W[i]);
At[i] = Zt[i].Func(activationFunc[i]);
}
else if (i == layers.Length - 1)
{
Zt[i] = At[i - 1].AddFeatureBias(1).Mul(W[i]);
At[i] = Zt[i].Func((x) => x);
}
else
{
Zt[i] = At[i - 1].AddFeatureBias(1).Mul(W[i]);
At[i] = Zt[i].Func(activationFunc[i]);
}
}
return(At[layers.Length - 2]);
}
private void Initialize()
{
X = layers[0].matrix.GetBatch(_batchSize, 0);
X = X.AddFeatureBias(1);
Y = AllY.GetBatch(_batchSize, 0);
W = new double[layers.Length - 1][, ];
Gradient = new double[layers.Length - 1][, ];
Sigma = new double[layers.Length - 1][, ];
A = new double[layers.Length - 1][, ];
Z = new double[layers.Length - 1][, ];
activationFunc = new Func <double, double> [layers.Length - 1];
activationFuncPrime = new Func <double, double> [layers.Length - 1];
for (int i = 1; i < layers.Length; i++)
{
if (i == 1)
{
A[i - 1] = new double[X.GetLength(0), layers[i].neurons];
W[i - 1] = new double[X.GetLength(1) - 1, layers[i].neurons];
}
else if (i == layers.Length - 1)
{
A[i - 1] = new double[layers[i - 1].neurons, layers[i].matrix.GetLength(1)];
W[i - 1] = new double[layers[i - 1].neurons, layers[i].matrix.GetLength(1)];
}
else
{
A[i - 1] = new double[layers[i - 1].neurons, layers[i].neurons];
W[i - 1] = new double[layers[i - 1].neurons, layers[i].neurons];
}
ActivationFunctionFactory.SetFunctions(ref activationFunc[i - 1], ref activationFuncPrime[i - 1], layers[i].Type);
W[i - 1] = W[i - 1].Rand().AddWeightBias(0.1);
}
}
