public List <Particle> Sample(int sampleCount, List <Particle> Particles)
{
// make N subdivisions, choose positions
// with a consistent random offset
var positions = Enumerable.Range(0, Particles.Count).Select(_ => (_ + SingleRandom.Instance.NextDouble()) / Particles.Count).ToArray();
//indexes = np.zeros(N, 'i')
var cumulative_sum = Particles.Select(_ => _.Weight).CumulativeSum().ToArray();
int i = 0, j = 0;
int[] indexes = new int[Particles.Count];
while (i < Particles.Count)
{
if (positions[i] < cumulative_sum[j])
{
indexes[i] = j;
i += 1;
}
else
{
j += 1;
}
}
return(ReSampleHelper.ReSampleFromIndex(Particles, indexes));
}
public static bool DoResample(double effectiveCountMinRatio, IList <Particle> particles)
{
var effectiveCountRatio =
(double)ReSampleHelper._effectiveParticleCount(ReSampleHelper.GetNormalizedWeights(particles)) / particles.Count;
return(effectiveCountRatio > Single.Epsilon && effectiveCountRatio < effectiveCountMinRatio);
}
protected void NextEpoch(double effectiveCountMinRatio)
{
////resample if too few effective particles
//if (Neff(particles) < N / 2)
// particles = particleSampler.Sample(particles);
Particles = ReSampleHelper.DoResample(effectiveCountMinRatio, Particles) ?
ReSampler.Sample(Particles.Count, Particles) :
Particles;
Particles = Particles.Select(p => (Particle)p.Clone()).ToList();
}
/*
* Given the current state estimate X and weight vector w, resample a new set of
* states.We use the low-variance resampling algorithm from Thrun, Burgard, and Fox's
* "Probabilistic Robotics". By default it will keep the number of samples constant
*/
public List <Particle> Sample(int sampleCount, List <Particle> Particles)
{
var r = SingleRandom.Instance.NextDouble() / Particles.Count;
int[] indexes = new int[Particles.Count];
for (int i = 0; i < Particles.Count; i++)
{
var u = r + (i - 1) / Particles.Count;
double c = 0;
int j = 0;
while (c < u)
{
j++;
c = +Particles[i].Weight;
}
indexes[i] = j;
}
return(ReSampleHelper.ReSampleFromIndex(Particles, indexes));
}
