/// <summary>
/// Creates a deep copy of this network.
/// </summary>
/// <returns>A new instance of <see cref="NeuralNetwork"/> which has the same layers as this network.</returns>
public NeuralNetwork Clone()
{
DenseLayer[] clonnedLayers = new DenseLayer[layers.Length];
for (int k = 0; k < layers.Length; k++)
{
clonnedLayers[k] = layers[k].Clone();
}
return(new NeuralNetwork(clonnedLayers));
}
/// <summary>
/// Constructs a new instance of <see cref="NeuralNetwork"/> using topology and weights specified in
/// the <paramref name="networkInString"/> argument.
/// </summary>
/// <remarks>
/// The given string must have 2 lines, the first containing layers' sizes
/// separated by ',' sign, and the second containing weights also separated by ','.
/// </remarks>
/// <param name="networkInString">An instance of string containing the topology and weights of the network.
/// </param>
/// <returns>The new constructed neural network.</returns>
/// <exception cref="ArgumentException">
/// Thrown if input parameter <paramref name="networkInString"/> contains invalid network representation.
/// </exception>
public static NeuralNetwork ConstructFromString(string networkInString)
{
string[] lines = networkInString.Split(new[] { "\r\n", "\r", "\n" }, StringSplitOptions.None);
if (lines.Length < 2)
{
throw new ArgumentException(string.Format("Invalid input parameter {0}. The string must contain 2 lines.", "networkInString"));
}
string[] sizes = lines[0].Split(',');
if (sizes.Length < 4)
{
throw new ArgumentException(string.Format("Invalid input parameter {0}. The string must contain sizes for at least 2 layers.", "networkInString"));
}
string[] weightsInString = lines[1].Split(',');
DenseLayer[] denseLayers = new DenseLayer[sizes.Length / 2];
int k = 0;
for (int c = 0; c < sizes.Length; c += 2)
{
int selfNeuronsCount;
if (!int.TryParse(sizes[c], out selfNeuronsCount) || selfNeuronsCount < 1)
{
throw new ArgumentException(string.Format("Invalid input parameter {0}. The neural layer must contain at least 1 neuron.", "networkInString"));
}
int outputNeuronsCount;
if (!int.TryParse(sizes[c + 1], out outputNeuronsCount) || outputNeuronsCount < 1)
{
throw new ArgumentException(string.Format("Invalid input parameter {0}. Output vector of neural layer must contain at least 1 neuron.", "networkInString"));
}
double[,] weights = new double[selfNeuronsCount + 1, outputNeuronsCount];
for (int i = 0; i <= selfNeuronsCount; i++)
{
for (int j = 0; j < outputNeuronsCount; j++)
{
double weight;
if (k >= weightsInString.Length || !double.TryParse(weightsInString[k++], out weight))
{
throw new ArgumentException(string.Format("Invalid input parameter {0}. Weights count must match to typology.", "networkInString"));
}
weights[i, j] = weight;
}
}
denseLayers[c / 2] = new DenseLayer(weights, LayerActivationFunctions.Sigmoid);
}
return(new NeuralNetwork(denseLayers));
}