Resampler

class stonesoup.resampler.base.Resampler

Bases: Base

Resampler base class

Particle

class stonesoup.resampler.particle.SystematicResampler

Bases: Resampler

Traditional style resampler for particle filter. Calculates first random point in (0, 1/nparts], then calculates nparts points that are equidistantly distributed across the CDF. Complexity of order O(N) where N is the number of resampled particles.

resample(particles, nparts=None)

Resample the particles

Parameters:

• particles (ParticleState or list of Particle) – The particles or particle state to be resampled according to their weights

• nparts (int) – The number of particles to be returned from resampling

Returns:

particle state – The particle state after resampling

Return type:

ParticleState

class stonesoup.resampler.particle.ESSResampler(threshold: float = None, resampler: Resampler = SystematicResampler())

Bases: Resampler

This wrapper uses a Resampler to resample the particles inside an instance of Particles, but only after checking if this is necessary by comparing the Effective Sample Size (ESS) with a supplied threshold (numeric). Kish’s ESS is used, as recommended in Section 3.5 of this tutorial [1].

References

[1]

Doucet A., Johansen A.M., 2009, Tutorial on Particle Filtering and Smoothing: Fifteen years later, Handbook of Nonlinear Filtering, Vol. 12.

Parameters:

• threshold (float, optional) – Threshold compared with ESS to decide whether to resample. Default is number of particles divided by 2, set in resample method

• resampler (Resampler, optional) – Resampler to wrap, which is called when ESS below threshold

threshold: float

Threshold compared with ESS to decide whether to resample. Default is number of particles divided by 2, set in resample method

resampler: Resampler

Resampler to wrap, which is called when ESS below threshold

resample(particles, nparts=None)

Parameters:

• particles (ParticleState or list of Particle) – The particles to be resampled according to their weight

• nparts (int) – The number of particles to be returned from resampling

Returns:

particles – The particles, either unchanged or resampled, depending on weight degeneracy

Return type:

ParticleState

class stonesoup.resampler.particle.MultinomialResampler

Bases: Resampler

Traditional style resampler for particle filter. Calculates a random point in (0, 1] individually for each particle, and picks the corresponding particle from the CDF calculated from particle weights. Complexity is of order O(NM) where N and M are the number of resampled and existing particles respectively.

resample(particles, nparts=None)

Resample the particles

Parameters:

• particles (ParticleState or list of Particle) – The particles or particle state to be resampled according to their weights

• nparts (int) – The number of particles to be returned from resampling

Returns:

particle state – The particle state after resampling

Return type:

ParticleState

class stonesoup.resampler.particle.StratifiedResampler

Bases: Resampler

Traditional style resampler for particle filter. Splits the CDF into N evenly sized subpopulations (‘strata’), then independently picks one value from each stratum. Complexity of order O(N).

resample(particles, nparts=None)

Resample the particles

Parameters:

• particles (ParticleState or list of Particle) – The particles or particle state to be resampled according to their weights

• nparts (int) – The number of particles to be returned from resampling

Returns:

particle state – The particle state after resampling

Return type:

ParticleState

class stonesoup.resampler.particle.ResidualMethod(value, names=None, *values, module=None, qualname=None, type=None, start=1, boundary=None)

Bases: Enum

class stonesoup.resampler.particle.ResidualResampler(residual_method: ResidualMethod = ResidualMethod.MULTINOMIAL)

Bases: Resampler

Wrapper around a traditional resampler. Any particle, p with weight W >= 1/N, will be resampled floor(W_p) times, providing N_stage_1 = sum(floor(W_p)) resampled particles.

The residual weights of each particle are carried over and passed into another resampler, where the remaining N - N_stage_1 particles are resampled from.

Should be a more computationally efficient method than resampling all particles from a CDF. Cannot be used to upsample or downsample.

Parameters:

residual_method (ResidualMethod, optional) – Method used to resample particles from the residuals.

residual_method: ResidualMethod

Method used to resample particles from the residuals.

resample(particles, nparts=None)

Resample the particles. For ResidualResampler, nparts must equal len(particles)

Parameters:

• particles (ParticleState or list of Particle) – The particles or particle state to be resampled according to their weights

• nparts (int) – The number of particles to be returned from resampling - must equal number of particles from previous step

Returns:

particle state – The particle state after resampling

Return type:

ParticleState