public NetworkState(RecurrentNetwork network, NetworkState last)
{
this.Network = network;
states = new List <object>();
double[] input = new double[network.Layers[0].InputSize];
this.Input = input;
double[] output = null;
for (int i = 0; i < network.Layers.Count; i++)
{
Layer layer = network.Layers[i];
output = new double[layer.Size];
if (layer is RecurrentLayer)
{
RecurrentState lastState = last != null ? (RecurrentState)last.states[i] : null;
states.Add(new RecurrentState(input, output, lastState));
}
else
{
states.Add(new FeedForwardState(input, output));
}
// The input of the next layer is the output of the last.
input = output;
}
this.Output = output;
}
public NetworkTrainingState(RecurrentNetwork network)
{
Network = network;
double[] inputErrors = new double[network.Layers[0].InputSize];
double[] errors = null;
states = new List <TrainingState>();
for (int i = 0; i < network.Layers.Count; i++)
{
Layer layer = network.Layers[i];
errors = new double[layer.Size];
if (layer is RecurrentLayer)
{
//RecurrentState lastState = last != null ? (RecurrentState)last.states[i] : null;
states.Add(new RecurrentTrainingState(inputErrors, errors));
}
else
{
states.Add(new FeedForwardTrainingState(inputErrors, errors));
}
// The input errors next layer are the errors of the last.
inputErrors = errors;
}
this.Errors = errors;
}
public NetworkState(RecurrentNetwork network)
: this(network, null)
{
}
public RecurrentNetworkTrainer(RecurrentNetwork network)
{
this.Network = network;
this.trainingState = new NetworkTrainingState(network);
}