public static void CompareGenomes(NeatGenome <double> x, NeatGenome <double> y)
{
// Compare connections.
var xGenes = x.ConnectionGenes;
var yGenes = y.ConnectionGenes;
ArrayTestUtils.Compare(xGenes._connArr, yGenes._connArr);
ArrayTestUtils.Compare(xGenes._weightArr, yGenes._weightArr);
// Compare hidden node ID arrays.
ArrayTestUtils.Compare(x.HiddenNodeIdArray, y.HiddenNodeIdArray);
}
public void TestCalcGradients()
{
const int sampleCount = 100;
ParamSamplingInfo psi = new ParamSamplingInfo(0, 2 * Math.PI, sampleCount);
double[] yArr = new double[sampleCount];
FuncRegressionUtils.Probe((x) => Math.Sin(x), psi, yArr);
// Calc gradients.
double[] gradientArr = new double[sampleCount];
FuncRegressionUtils.CalcGradients(psi, yArr, gradientArr);
// Calc expected gradients (using simple non-vectorized logic).
double[] gradientArrExpected = new double[sampleCount];
CalcGradients(psi, yArr, gradientArrExpected);
// Compare results.
ArrayTestUtils.Compare(gradientArrExpected, gradientArr);
}
public void TestAddAcyclicConnection_CumulativeAdditions()
{
var pop = CreateNeatPopulation();
var generationSeq = new Int32Sequence();
var genomeBuilder = NeatGenomeBuilderFactory <double> .Create(pop.MetaNeatGenome);
var rootGenome = pop.GenomeList[0];
var strategy = new AddAcyclicConnectionStrategy <double>(
pop.MetaNeatGenome, genomeBuilder,
pop.GenomeIdSeq, pop.InnovationIdSeq, generationSeq);
IRandomSource rng = RandomDefaults.CreateRandomSource();
var nodeIdSet = GetNodeIdSet(rootGenome);
CyclicGraphAnalysis cyclicGraphAnalysis = new CyclicGraphAnalysis();
AcyclicGraphDepthAnalysis graphDepthAnalysis = new AcyclicGraphDepthAnalysis();
// Run the inner loop test multiple times.
// Note. The add-connection mutations are random, thus each loop accumulates a different set of mutations.
for (int i = 0; i < 50; i++)
{
var parentGenome = rootGenome;
// Accumulate random mutations for some number of loops.
for (int j = 0; j < 20;)
{
var childGenome = strategy.CreateChildGenome(rootGenome, rng);
// Note. the strategy will return a null if it cannot find an acyclic connection to add;
// test for this and try again. The test will be for N successful mutations rather than N attempts.
if (null == childGenome)
{
continue;
}
// The child genome should have one more connection than parent.
Assert.AreEqual(rootGenome.ConnectionGenes.Length + 1, childGenome.ConnectionGenes.Length);
// The child genome's new connection should not be a duplicate of any of the existing/parent connections.
var connSet = GetDirectedConnectionSet(rootGenome);
var childConnSet = GetDirectedConnectionSet(childGenome);
var newConnList = new List <DirectedConnection>(childConnSet.Except(connSet));
Assert.AreEqual(1, newConnList.Count);
// The connection genes should be sorted.
Assert.IsTrue(SortUtils.IsSortedAscending(childGenome.ConnectionGenes._connArr));
// The child genome should have the same set of node IDs as the parent.
var childNodeIdSet = GetNodeIdSet(childGenome);
Assert.IsTrue(nodeIdSet.SetEquals(childNodeIdSet));
// The child genome should describe an acyclic graph, i.e. the new connection should not have
// formed a cycle in the graph.
var digraph = childGenome.DirectedGraph;
Assert.IsFalse(cyclicGraphAnalysis.IsCyclic(digraph));
// Run the acyclic graph depth analysis algorithm.
GraphDepthInfo depthInfo = graphDepthAnalysis.CalculateNodeDepths(childGenome.DirectedGraph);
// Run again with the alternative algorithm (that uses function recursion).
GraphDepthInfo depthInfo2 = AcyclicGraphDepthAnalysisByRecursion.CalculateNodeDepths(childGenome.DirectedGraph);
Assert.AreEqual(nodeIdSet.Count, depthInfo._nodeDepthArr.Length);
Assert.AreEqual(nodeIdSet.Count, depthInfo2._nodeDepthArr.Length);
ArrayTestUtils.Compare(depthInfo2._nodeDepthArr, depthInfo._nodeDepthArr);
// Set the child genome to be the new parent, thus we accumulate random new connections over time.
parentGenome = childGenome;
// Increment for successful tests only.
j++;
}
}
}