State of the Particle Filter
Advance the system of particles to the next timepoint
Advance the system of particles to the next timepoint
the system evolution matrix, G_t
the time of the next observation
a collection of particles representing time emprical posterior distribution of the state at time - 1
the parameters of the model, containing the system evolution matrix, W
a distribution over a collection of particles at this time
Calculate the weights of each particle using the conditional likelihood of the observations given the state
Calculate the weights of each particle using the conditional likelihood of the observations given the state
the DGLM model specification containing F_t and G_t, and the conditional likelihood of observations
the time of the observation
a collection of particles representing time emprical prior distribution of the state at this time
the observation observation given a value of the state, p(y_t | F_t x_t)
a collection of weights corresponding to each particle
Run a Boostrap Particle Filter over a DGLM
Run a Boostrap Particle Filter over a DGLM
a Model class containing the specification of G and F
a Seq of Data
the parameters used to run the particle filter
the number of particles to use in the filter, more results in higher accuracy
Run a Bootstrap Particle Filter over a DGLM to calculate the log-likelihood
Multinomial Resampling
The particle filter can be used for inference of Dynamic Generalised Linear Models (DGLMs), where the observation distribution is not Gaussian.