Using a Particle Filter

Particle filters can help you find probable states of your network

Previous page Next page


Particle filters can help you find probable states of your network which can be used as starting states for MCMC sampling. Here is an example on how to use Keanu’s Particle filter class:

//Create a dummy Bayesian Network
DoubleVertex temperature = new UniformVertex(0.0, 100.0);
DoubleVertex noiseAMu = new GaussianVertex(0.0, 2.0);
DoubleVertex noiseA = new GaussianVertex(noiseAMu, 2.0);
DoubleVertex noiseBMu = new GaussianVertex(0.0, 2.0);
DoubleVertex noiseB = new GaussianVertex(noiseBMu, 2.0);
DoubleVertex thermometerA = new GaussianVertex(, 1.0);
DoubleVertex thermometerB = new GaussianVertex(, 1.0);

//Create a particle filter with default settings
ParticleFilter filter = ParticleFilter.ofVertexInGraph(temperature)

//Get a sorted list of the most probable particles in order of descending probability
List<Particle> particles = filter.getSortedMostProbableParticles();

//Get the most probable particle
Particle mostProbableParticle = filter.getMostProbableParticle();

//Get the estimated temperature in the most probable particle
double estimation = mostProbableParticle.getScalarValueOfVertex(temperature);

//Get the probability of this particle state
double logProb = mostProbableParticle.logProb();