Hidden state models
HiddenStateModel{S, T<:TimeType} <: AbstractModel{S}
Abstract type for any model of the hidden state of type S
.
FeedbackParticleFilters.state_dim
— Function.state_dim(problem::AbstractFilteringProblem)
Return the dimensionality of the hidden state in problem
.
state_dim(model::HiddenStateModel)
Returns the dimensionality of the hidden state in model
.
FeedbackParticleFilters.state_type
— Function.state_type(arg)
Returns the type of the hidden state in arg
. Supported argument types:
- Any subtype of AbstractModel
- Any subtype of AbstractFilteringProblem
state_type(model::HiddenStateModel)
Returns the data type of the hidden state in model
.
Missing docstring for time_type
. Check Documenter's build log for details.
FeedbackParticleFilters.initial_condition
— Function.initial_condition(model::HiddenStateModel)
Returns the specification of the initial condition in model
. This is either a fixed value or a samplable distribution.
FeedbackParticleFilters.initialize
— Function.initialize(model::HiddenStateModel)
Returns a sample from the initial distribution of model
.
FeedbackParticleFilters.propagate!
— Function.propagate!(state[s], model::HiddenStateModel[, dt])
Propagates the state(s) according to the model. For ContinuousTime' models, a time step
dt' has to be provided. Multiple states are given as a matrix with columns corresponding to states, and are processed i.i.d.
propagate!(sfs, filtering_problem, filtering_algorithm; dt) --> sfs
Propagates the system and filter states for one time-step according to the specified filtering problem and algorithm.
Diffusion processes
Missing docstring for DiffusionStateModel
. Check Documenter's build log for details.
LinearDiffusionStateModel(A, B; init)
Returns a linear diffusion process hidden state model $dX_t = A X_t dt + B dW_t$ with appropriately sized matrices $A$ and $B$.
Optional argument init
stands for the initial condition of the process, which is either
- A vector of length
n
for a fixed (deterministic) initial condition - A
Distributions.Sampleable
type for a random initial condition
If argument init
is left out, it is set to either
- a multivariate normal distribution with covariance matrix set to the stationary variance, if it exists
- the zero vector
Missing docstring for ScalarDiffusionStateModel
. Check Documenter's build log for details.
FeedbackParticleFilters.drift_function
— Function.drift_function(model)
Returns the drift function $f$ of the diffusion model $dX_t = f(X_t)dt + g(X_t)dW_t$.
FeedbackParticleFilters.diffusion_function
— Function.diffusion_function(model)
Returns the diffusion function $g$ of the diffusion model $dX_t = f(X_t)dt + g(X_t)dW_t$.
FeedbackParticleFilters.noise_dim
— Function.noise_dim(model)
Returns the dimension of the Brownian motion $W_t$ in the diffusion model $dX_t = f(X_t)dt + g(X_t)dW_t$.