public void ShouldValidate(bool?input, OptimizationModes expected)
{
var target = new DisableBailInput {
SuppliedInput = input
};
var result = target.Validate();
result.ShouldBe(expected);
}
public void init(int dim, int populationSize, double[] initialMean, float initialStepSize, OptimizationModes mode)
{
this.dim = dim;
this.mode = mode;
int N = dim;
lambda = populationSize; // population size, e.g., 4+floor(3*log(N));
mu = lambda / 2; // number of parents, e.g., floor(lambda/2);
sigma = 0.3 * 2.0f; // initial step-size
chiN = Math.Sqrt((double)N) * (1.0 - 1.0 / (4.0 * N) + 1 / (21.0 * N * N));
M = new double[N, N];
Mmult = new double[N, N];
ps = new double[N];
xmean = new double[N];
xBest = new double[N];
zmean = new double[N];
xold = new double[N];
z = new double[N];
Mz = new double[N];
weights = new double[mu];
arfitness = new double[lambda];
arindex = new int[lambda];
arr_tmp = new List<SortedVal>(lambda);
bestFitness = mode == OptimizationModes.minimize ? double.PositiveInfinity : double.NegativeInfinity;
for (int i = 0; i < lambda; i++)
{
arx.Add(new double[N]);
arz.Add(new double[N]);
ard.Add(new double[N]);
}
for (int i = 0; i < lambda; i++)
{
arr_tmp.Add(new SortedVal());
}
for (int i = 0; i < N; i++)
for (int j = 0; j < N; j++)
if (i == j) M[i, j] = 1.0;
else M[i, j] = 0.0;
double sum_weights = 0;
for (int i = 0; i < mu; i++)
{
weights[i] = Math.Pow(Math.Log(0.5 + (double)mu) - Math.Log((double)(1 + i)), 1.0);
sum_weights = sum_weights + weights[i];
}
mueff = 0;
for (int i = 0; i < mu; i++)
{
weights[i] = weights[i] / sum_weights;
mueff = mueff + weights[i] * weights[i];
}
mueff = 1.0 / mueff;
for (int i = 0; i < N; i++)
{
ps[i] = 0;
}
cs = (mueff + 2.0) / (N + mueff + 5.0);
lcoef = 2.0;
c1 = lcoef / (Math.Pow(N + 1.3, 2.0) + mueff);
cmu = minv(1.0 - c1, lcoef * (mueff - 2.0 + 1.0 / mueff) / (Math.Pow(N + 2.0, 2.0) + mueff));
if (initialMean.Length != dim)
Debug.LogException(new Exception("Incorrect array length!"));
Array.Copy(initialMean, xmean, xmean.Length);
sigma = initialStepSize;
}
public void init(int dim, int populationSize, double[] initialMean, float initialStepSize, OptimizationModes mode)
{
this.dim = dim;
this.mode = mode;
int N = dim;
// set boring parameters
lambda = populationSize;
mu = lambda / 2; // number of parents, e.g., floor(lambda/2);
sigma = initialStepSize; // initial step-size
chiN = Math.Sqrt((double)N) * (1.0 - 1.0 / (4.0 * N) + 1 / (21.0 * N * N));
// set interesting (to be tuned) parameters
nvectors = 4 + 3 * (int)Math.Log((double)N); // number of stored direction vectors, e.g., nvectors = 4+floor(3*log(N))
cs = 2.0 * ((double)lambda) / N; // nvectors;
// memory allocation
//ibit = 31; // for the fast Rademacher sampling
arx = new List<double[]>();
arz = new List<double[]>();
for (int i = 0; i < lambda; i++)
{
arx.Add(new double[N]);
arz.Add(new double[N]);
}
ps_arr = new double[nvectors, N];
ps = new double[N];
xmean = new double[N];
xBest = new double[N];
zmean = new double[N];
xold = new double[N];
z = new double[N];
dz = new double[N];
Az = new double[N];
weights = new double[mu];
arfitness = new double[lambda];
arindex = new int[lambda];
arr_tmp = new List<SortedVal>(lambda);
for (int i = 0; i < lambda; i++)
{
arr_tmp.Add(new SortedVal());
}
double sum_weights = 0;
for (int i = 0; i < mu; i++)
{
weights[i] = Math.Pow(Math.Log(0.5 + (double)mu) - Math.Log((double)(1 + i)), 1.0);
sum_weights = sum_weights + weights[i];
}
mueff = 0;
for (int i = 0; i < mu; i++)
{
weights[i] = weights[i] / sum_weights;
mueff = mueff + weights[i] * weights[i];
}
mueff = 1.0 / mueff;
if (N < 32)
Debug.LogWarning("Warning: LM-MA-ES is not optimal for low-dimensional problems. You should use MA-ES instead");
//With small N, LMMAES way of computing cs can result in negative or zero values.
//In this case, we revert to how it's computed in CMA-ES and MA-ES.
//However, this is a hack to prevent crashes, and one should rather use MA-ES than LM-MA-ES with small N.
cs = (mueff + 2.0) / (N + mueff + 5.0);
for (int i = 0; i < N; i++)
{
ps[i] = 0;
}
if (initialMean.Length != dim)
Debug.LogException(new Exception("Incorrect array length!"));
Array.Copy(initialMean, xmean, xmean.Length);
for (int i = 0; i < nvectors; i++)
for (int j = 0; j < N; j++)
ps_arr[i, j] = 0;
bestFitness = mode == OptimizationModes.minimize ? double.PositiveInfinity : double.NegativeInfinity;
iterator_sz = 0;
}
