Edit on GitHub

DPOMP models

Discrete-state space Partially Observed Markov Processes

Predefined models

The package includes the following (mostly) epidemiological models as predefined examples:

  • "SI"
  • "SIR"
  • "SIS"
  • "SEI"
  • "SEIR"
  • "SEIS"
  • "SEIRS"
  • "PREDPREY"
  • "ROSSMAC"

They can be instantiated using the generate_model function:

using BayesianWorkflows
model_name = "SIS"
initial_model_state = [100, 1]
model = generate_model(model_name, initial_model_state)

NB. this generates a standard configuration of the named model that can be tweaked later.

Options

Aside from the model name and initial model state (i.e. the initial 'population',) there are a number of options for generating a default model configuration:

  1. freq_dep – epidemiological models only, set to true for frequency-dependent contact rates.
  2. obs_error – average observation error (default = 2.)
  3. t0_index – index of the parameter that represents the initial time. 0 if fixed at 0.0.

Full model configuration

BayesianWorkflows.DPOMPModel — Type
DPOMPModel

A mutable struct which represents a DSSCT model (see Models for further details).

Fields

  • name – string, e,g, "SIR".
  • n_events – number of distinct event types.
  • event_rates! – event rate function.
  • initial_state – function for sampling initial model (i.e. population) state.
  • transition! – transition function of form f!(population, event_type).
  • obs_model – observation model likelihood function.
  • obs_function – observation function, use this to add 'noise' to simulated observations.
  • t0_index – index of the parameter that represents the initial time. 0 if fixed at 0.0.
source