/// <summary>
/// Constructs a cyclic neural network.
/// </summary>
/// <param name="digraph">The directed graph that defines the neural network structure.</param>
/// <param name="weightArr">An array of neural network connection weights.</param>
/// <param name="activationFn">The neuron activation function to use at all neurons in the network.</param>
/// <param name="cyclesPerActivation">The number of activation cycles to perform per overall activation of the cyclic network.</param>
public NeuralNetCyclic(
DirectedGraph digraph,
double[] weightArr,
VecFn2 <double> activationFn,
int cyclesPerActivation)
{
Debug.Assert(digraph.ConnectionIdArrays._sourceIdArr.Length == weightArr.Length);
// Store refs to network structure data.
_srcIdArr = digraph.ConnectionIdArrays._sourceIdArr;
_tgtIdArr = digraph.ConnectionIdArrays._targetIdArr;
_weightArr = weightArr;
// Store network activation function and parameters.
_activationFn = activationFn;
_cyclesPerActivation = cyclesPerActivation;
// Store input/output node counts.
_inputCount = digraph.InputCount;
_outputCount = digraph.OutputCount;
// Create node pre- and post-activation signal arrays.
int nodeCount = digraph.TotalNodeCount;
_preActivationArr = new double[nodeCount];
_postActivationArr = new double[nodeCount];
// Wrap sub-ranges of the neuron signal arrays as input and output vectors.
this.InputVector = new VectorSegment <double>(_postActivationArr, 0, _inputCount);
// Note. Output neurons follow input neurons in the arrays.
this.OutputVector = new VectorSegment <double>(_postActivationArr, _inputCount, _outputCount);
}
/// <summary>
/// Constructs a cyclic neural network.
/// </summary>
/// <param name="digraph">The directed graph that defines the neural network structure.</param>
/// <param name="weightArr">An array of neural network connection weights.</param>
/// <param name="activationFn">The neuron activation function to use at all neurons in the network.</param>
/// <param name="cyclesPerActivation">The number of activation cycles to perform per overall activation of the cyclic network.</param>
public NeuralNetCyclic(
DirectedGraph digraph,
double[] weightArr,
VecFn2 <double> activationFn,
int cyclesPerActivation)
{
Debug.Assert(digraph.ConnectionIds.GetSourceIdSpan().Length == weightArr.Length);
// Store refs to network structure data.
_connIds = digraph.ConnectionIds;
_weightArr = weightArr;
// Store network activation function and parameters.
_activationFn = activationFn;
_cyclesPerActivation = cyclesPerActivation;
// Store input/output node counts.
_inputCount = digraph.InputCount;
_outputCount = digraph.OutputCount;
_totalNodeCount = digraph.TotalNodeCount;
// Get a working array for both pre and post node activation signals, and map memory segments to pre
// and post signal segments.
// Rent an array that has length of at least _totalNodeCount * 2.
_activationsArr = ArrayPool <double> .Shared.Rent(_totalNodeCount << 1);
_preActivationMem = _activationsArr.AsMemory(0, _totalNodeCount);
_postActivationMem = _activationsArr.AsMemory(_totalNodeCount, _totalNodeCount);
// Map the input and output vectors to the corresponding segments of _postActivationArr.
this.Inputs = _postActivationMem.Slice(0, _inputCount);
// Note. Output neurons follow input neurons in the arrays.
this.Outputs = _postActivationMem.Slice(_inputCount, _outputCount);
}
/// <summary>
/// Constructs a cyclic neural network.
/// </summary>
/// <param name="digraph">The directed graph that defines the neural network structure.</param>
/// <param name="weightArr">An array of neural network connection weights.</param>
/// <param name="activationFn">The neuron activation function to use at all neurons in the network.</param>
/// <param name="cyclesPerActivation">The number of activation cycles to perform per overall activation of the cyclic network.</param>
public NeuralNetCyclicSafe(
DirectedGraph digraph,
double[] weightArr,
VecFn2 <double> activationFn,
int cyclesPerActivation)
{
Debug.Assert(digraph.ConnectionIds.GetSourceIdSpan().Length == weightArr.Length);
// Store refs to network structure data.
_connIds = digraph.ConnectionIds;
_weightArr = weightArr;
// Store network activation function and parameters.
_activationFn = activationFn;
_cyclesPerActivation = cyclesPerActivation;
// Store input/output node counts.
_inputCount = digraph.InputCount;
_outputCount = digraph.OutputCount;
// Create node pre- and post-activation signal arrays.
int nodeCount = digraph.TotalNodeCount;
_preActivationArr = new double[nodeCount];
_postActivationArr = new double[nodeCount];
// Map the input and output vectors to the corresponding segments of _postActivationArr.
this.Inputs = new Memory <double>(_postActivationArr, 0, _inputCount);
// Note. Output neurons follow input neurons in the arrays.
this.Outputs = new Memory <double>(_postActivationArr, _inputCount, _outputCount);
}
private void RunBenchmark(IActivationFunction <double> actFn)
{
VecFn2 <double> fn = actFn.Fn;
for (int i = 0; i < __loops; i++)
{
fn(_x, _w);
}
}
/// <summary>
/// Constructs a cyclic neural network.
/// </summary>
/// <param name="digraph">The weighted directed graph that defines the neural network structure and connection weights.</param>
/// <param name="activationFn">The neuron activation function to use at all neurons in the network.</param>
/// <param name="cyclesPerActivation">The number of activation cycles to perform per overall activation of the cyclic network.</param>
public NeuralNetCyclic(
WeightedDirectedGraph <double> digraph,
VecFn2 <double> activationFn,
int cyclesPerActivation)
: this(
digraph,
digraph.WeightArray,
activationFn,
cyclesPerActivation)
{
}
