Skip to contents

Likelihood-Free Markhov Chain Monte Carlo (LFMCMC)

Source: R/LFMCMC.R
LFMCMC.Rd

Likelihood-Free Markhov Chain Monte Carlo (LFMCMC)

Usage

LFMCMC(model = NULL)

run_lfmcmc(lfmcmc, params_init, n_samples, epsilon, seed = NULL)

set_observed_data(lfmcmc, observed_data)

set_proposal_fun(lfmcmc, fun)

use_proposal_norm_reflective(lfmcmc)

set_simulation_fun(lfmcmc, fun)

set_summary_fun(lfmcmc, fun)

set_kernel_fun(lfmcmc, fun)

use_kernel_fun_gaussian(lfmcmc)

get_mean_params(lfmcmc)

get_mean_stats(lfmcmc)

get_initial_params(lfmcmc)

get_current_proposed_params(lfmcmc)

get_current_accepted_params(lfmcmc)

get_current_proposed_stats(lfmcmc)

get_current_accepted_stats(lfmcmc)

get_observed_stats(lfmcmc)

get_all_sample_params(lfmcmc)

get_all_sample_stats(lfmcmc)

get_all_sample_acceptance(lfmcmc)

get_all_sample_drawn_prob(lfmcmc)

get_all_sample_kernel_scores(lfmcmc)

get_all_accepted_params(lfmcmc)

get_all_accepted_stats(lfmcmc)

get_all_accepted_kernel_scores(lfmcmc)

get_n_samples(lfmcmc)

get_n_stats(lfmcmc)

get_n_params(lfmcmc)

# S3 method for class 'epiworld_lfmcmc'
verbose_off(x)

set_params_names(lfmcmc, names)

set_stats_names(lfmcmc, names)

# S3 method for class 'epiworld_lfmcmc'
print(x, burnin = 0, ...)

Arguments

model

A model of class epiworld_model or NULL (see details).

lfmcmc

LFMCMC model

params_init

Initial model parameters, treated as double

n_samples

Number of samples, treated as integer

epsilon

Epsilon parameter, treated as double

seed

Random engine seed

observed_data

Observed data, treated as double.

fun

A function (see details).

x

LFMCMC model to print

names

Character vector of names.

burnin

Integer. Number of samples to discard as burnin before computing the summary.

...

Ignored

Value

The LFMCMC function returns a model of class epiworld_lfmcmc.

The simulated model of class epiworld_lfmcmc.

  • use_kernel_fun_gaussian: The LFMCMC model with kernel function set to gaussian.

  • get_mean_params: The param means for the given lfmcmc model.

  • get_mean_stats: The stats means for the given lfmcmc model.

  • The function get_initial_params returns the initial parameters for the given LFMCMC model.

  • The function get_current_proposed_params returns the proposed parameters for the next LFMCMC sample.

  • The function get_current_accepted_params returns the most recently accepted parameters (the current state of the LFMCMC)

  • The function get_current_proposed_stats returns the statistics from the simulation run with the proposed parameters

  • The function get_current_accepted_stats returns the statistics from the most recently accepted parameters

  • The function get_observed_stats returns the statistics for the observed data

  • The function get_all_sample_params returns a matrix of sample parameters for the given LFMCMC model. with the number of rows equal to the number of samples and the number of columns equal to the number of parameters.

  • The function get_all_sample_stats returns a matrix of statistics for the given LFMCMC model. with the number of rows equal to the number of samples and the number of columns equal to the number of statistics.

  • The function get_all_sample_acceptance returns a vector of boolean flags which indicate whether a given sample was accepted

  • The function get_all_sample_drawn_prob returns a vector of drawn probabilities for each sample

  • The function get_all_sample_kernel_scores returns a vector of kernel scores for each sample

  • The function get_all_accepted_params returns a matrix of accepted parameters for the given LFMCMC model. with the number of rows equal to the number of samples and the number of columns equal to the number of parameters.

  • The function get_all_accepted_stats returns a matrix of accepted statistics for the given LFMCMC model. with the number of rows equal to the number of samples and the number of columns equal to the number of statistics.

  • The function get_all_accepted_kernel_scores returns a vector of kernel scores for each accepted sample

  • The functions get_n_samples, get_n_stats, and get_n_params return the number of samples, statistics, and parameters for the given LFMCMC model, respectively.

  • The verbose_on and verbose_off functions return the same model, however verbose_off returns the model with no progress bar.

  • set_params_names: The lfmcmc model with the parameter names added.

  • set_stats_names: The lfmcmc model with the stats names added.

Details

Performs a Likelihood-Free Markhov Chain Monte Carlo simulation. When model is not NULL, the model uses the same random-number generator engine as the model. Otherwise, when model is NULL, a new random-number generator engine is created.

The functions passed to the LFMCMC object have different arguments depending on the object:

  • set_proposal_fun: A vector of parameters and the model.

  • set_simulation_fun: A vector of parameters and the model.

  • set_summary_fun: A vector of simulated data and the model.

  • set_kernel_fun: A vector of simulated statistics, observed statistics, epsilon, and the model.

The verbose_on and verbose_off functions activate and deactivate printing progress on screen, respectively. Both functions return the model (x) invisibly.

Examples

## Setup an SIR model to use in the simulation
model_seed <- 122
model_sir <- ModelSIR(name = "COVID-19", prevalence = .1,
  transmission_rate = .9, recovery_rate = .3)
agents_smallworld(
  model_sir,
  n = 1000,
  k = 5,
  d = FALSE,
  p = 0.01
)
verbose_off(model_sir)
run(model_sir, ndays = 50, seed = model_seed)

## Setup LFMCMC
# Extract the observed data from the model
obs_data <- get_today_total(model_sir)

# Define the simulation function
simfun <- function(params, lfmcmc_obj) {
  set_param(model_sir, "Recovery rate", params[1])
  set_param(model_sir, "Transmission rate", params[2])
  run(model_sir, ndays = 50)
  res <- get_today_total(model_sir)
  return(res)
}

# Define the summary function
sumfun <- function(dat, lfmcmc_obj) {
  return(dat)
}

# Create the LFMCMC model
lfmcmc_model <- LFMCMC(model_sir) |>
  set_simulation_fun(simfun) |>
  set_summary_fun(sumfun) |>
  use_proposal_norm_reflective() |>
  use_kernel_fun_gaussian() |>
  set_observed_data(obs_data)

## Run LFMCMC simulation
# Set initial parameters
par0 <- c(0.1, 0.5)
n_samp <- 2000
epsil <- 1.0

# Run the LFMCMC simulation
verbose_off(lfmcmc_model)
run_lfmcmc(
  lfmcmc = lfmcmc_model,
  params_init = par0,
  n_samples = n_samp,
  epsilon = epsil,
  seed = model_seed
)

# Print the results
set_stats_names(lfmcmc_model, get_states(model_sir))
set_params_names(lfmcmc_model, c("Immune recovery", "Infectiousness"))

print(lfmcmc_model)
#> ___________________________________________
#> 
#> LIKELIHOOD-FREE MARKOV CHAIN MONTE CARLO
#> 
#> N Samples (total) : 2000
#> N Samples (after burn-in period) : 2000
#> Elapsed t : 1.00s
#> 
#> Parameters:
#>   -Immune recovery :  0.45 [ 0.14,  0.95] (initial :  0.10)
#>   -Infectiousness  :  0.85 [ 0.54,  1.00] (initial :  0.50)
#> 
#> Statistics:
#>   -Susceptible :     0.22 [    0.00,     2.00] (Observed:     0.00)
#>   -Infected    :     0.07 [    0.00,     1.00] (Observed:     0.00)
#>   -Recovered   :   995.71 [  998.00,  1000.00] (Observed:  1000.00)
#> ___________________________________________
#> 

get_mean_stats(lfmcmc_model)
#> [1]   0.2215   0.0655 995.7130
get_mean_params(lfmcmc_model)
#> [1] 0.4482147 0.8480773