Package 'drjacoby'

Title: Flexible Markov Chain Monte Carlo via Reparameterization
Description: drjacoby is an R package for performing Bayesian inference via Markov chain monte carlo (MCMC). In addition to being highly flexible it implements some advanced techniques that can improve mixing in tricky situations.
Authors: Bob Verity [aut, cre], Pete Winskill [aut]
Maintainer: Bob Verity <[email protected]>
License: MIT + file LICENSE
Version: 1.5.4
Built: 2024-11-23 03:19:42 UTC
Source: https://github.com/mrc-ide/drjacoby

Help Index

Estimate autocorrelation

Description

Estimate autocorrelation

Usage

acf_data(x, lag)

Arguments

x

Single chain.
lag

maximum lag. Must be an integer between 1 and 500.

Check that drjacoby package has loaded successfully

Description

Simple function to check that drjacoby package has loaded successfully. Prints "drjacoby loaded successfully!" if so.

Usage

check_drjacoby_loaded()

Create template for cpp

Description

Create template for cpp

Usage

cpp_template(save_as)

Arguments

save_as

Path of (.cpp) file to create, relative to root of active project.

Define parameters dataframe

Description

Provides a convenient way of defining parameters in the format required by run_mcmc(). Each parameter must have the following three elements, defined in order:

  • name - the parameter name.

  • min - the minimum value of the parameter. -Inf is allowed.

  • max - the maximum value of the parameter. Inf is allowed.

There following arguments are also optional:

  • init - the initial value of the parameter. If running multiple chains a vector of initial values can be used to specify distinct values for each chain.

  • block - which likelihood block(s) this parameter belongs to. See vignettes for instructions on using likelihood blocks.

Usage

define_params(...)

Arguments

...

a series of named input arguments.

Examples

define_params(name = "mu", min = -10, max = 10, init = 0,
              name = "sigma", min = 0, max = 5, init = c(1, 2))
              
define_params(name = "mu1", min = -10, max = 10, init = 0, block = 1,
              name = "mu2", min = -10, max = 10, init = 0, block = 2,
              name = "sigma", min = 0, max = 5, init = 1, block = c(1, 2))

Flexible Markov Chain Monte Carlo via Reparameterization

Description

Flexible Markov chain monte carlo via reparameterization using the Jacobean matrix.

_PACKAGE

Gelman-Rubin statistic

Description

Estimate sthe Gelman-Rubin (rhat) convergence statistic for a single parameter across multiple chains. Basic method, assuming all chains are of equal length

Usage

gelman_rubin(par_matrix, chains, samples)

Arguments

par_matrix

Matrix (interations x chains)
chains

number of chains
samples

number of samples

Value

Gelman-Rubin statistic

References

Gelman, A., and D. B. Rubin. 1992. Inference from Iterative Simulation Using Multiple Sequences. Statistical Science 7: 457–511.

https://astrostatistics.psu.edu/RLectures/diagnosticsMCMC.pdf

Plot autocorrelation

Description

Plot autocorrelation for specified parameters

Usage

plot_autocorrelation(x, lag = 20, par = NULL, chain = 1, phase = "sampling")

Arguments

x

an object of class drjacoby_output
lag

maximum lag. Must be an integer between 1 and 500.
par

vector of parameter names. If NULL all parameters are plotted.
chain

which chain to plot.
phase

which phase to plot. Must be either "burnin" or "sampling".

Plot posterior correlation matrix

Description

Produces a matrix showing the correlation between all parameters from posterior draws.

Usage

plot_cor_mat(x, show = NULL, phase = "sampling", param_names = NULL)

Arguments

x

an object of class drjacoby_output
show

Vector of parameter names to plot.
phase

which phase to plot. Must be either "burnin" or "sampling".
param_names

Optional vector of names to replace the default parameter names.

Plot 95% credible intervals

Description

Plots posterior 95% credible intervals over specified set of parameters (defauls to all parameters).

Usage

plot_credible(x, show = NULL, phase = "sampling", param_names = NULL)

Arguments

x

an object of class drjacoby_output
show

vector of parameter names to plot.
phase

which phase to plot. Must be either "burnin" or "sampling".
param_names

optional vector of names to replace the default parameter names.

Produce density plots

Description

Density plots of all parameters. Use show and hide to be more specific about which parameters to plot.

Usage

plot_density(x, show = NULL, hide = NULL)

Arguments

x

an object of class drjacoby_output
show

optional vector of parameter names to plot. Parameters matching show will be included.
hide

optional vector of parameter names to filter out. Parameters matching hide will be hidden.

Plot Metropolis coupling acceptance rates

Description

Plot Metropolis coupling acceptance rates between all rungs.

Usage

plot_mc_acceptance(x, chain = NULL, phase = "sampling", x_axis_type = 1)

Arguments

x

an object of class drjacoby_output
chain

which chain to plot. If NULL then plot all chains.
phase

which phase to plot. Must be either "burnin" or "sampling".
x_axis_type

how to format the x-axis. 1 = integer rungs, 2 = values of the thermodynamic power.

Produce scatterplots between multiple parameters

Description

Uses ggpairs function from the GGally package to produce scatterplots between all named parameters.

Usage

plot_pairs(x, show = NULL, hide = NULL)

Arguments

x

an object of class drjacoby_output
show

optional vector of parameter names to plot. Parameters matching show will be included.
hide

optional vector of parameter names to filter out. Parameters matching hide will be hidden.

Plot loglikelihood 95% credible intervals

Description

Plot loglikelihood 95% credible intervals.

Usage

plot_rung_loglike(
  x,
  chain = 1,
  phase = "sampling",
  x_axis_type = 1,
  y_axis_type = 1
)

Arguments

x

an object of class drjacoby_output
chain

which chain to plot.
phase

which phase to plot. Must be either "burnin" or "sampling".
x_axis_type

how to format the x-axis. 1 = integer rungs, 2 = values of the thermodynamic power.
y_axis_type

how to format the y-axis. 1 = raw values, 2 = truncated at auto-chosen lower limit. 3 = double-log scale.

Produce bivariate scatterplot

Description

Produces scatterplot between two named parameters.

Usage

plot_scatter(
  x,
  parameter1,
  parameter2,
  downsample = TRUE,
  phase = "sampling",
  chain = NULL
)

Arguments

x

an object of class drjacoby_output
parameter1

name of parameter first parameter.
parameter2

name of parameter second parameter.
downsample

whether to downsample output to 200 values max to speed up plotting.
phase

which phase to plot. Must be either "burnin" or "sampling".
chain

which chain to plot.

Plot parameter trace

Description

Produce a series of plots corresponding to each parameter, including the raw trace, the posterior histogram and an autocorrelation plot. Plotting objects can be cycled through interactively, or can be returned as an object allowing them to be viewed/edited by the user.

Usage

plot_trace(
  x,
  show = NULL,
  hide = NULL,
  lag = 20,
  downsample = TRUE,
  phase = "sampling",
  chain = NULL,
  display = TRUE
)

Arguments

x

an object of class drjacoby_output
show

optional vector of parameter names to plot. Parameters matching show will be included.
hide

optional vector of parameter names to filter out. Parameters matching hide will be hidden.
lag

maximum lag. Must be an integer between 1 and 500.
downsample

boolean. Whether to downsample chain to make plotting more efficient.
phase

which phase to plot. Must be either "burnin", "sampling" or "both".
chain

which chain to plot.
display

boolean. Whether to show plots, if FALSE then plotting objects are returned without displaying.

Population data.

Description

Example population growth data.

Usage

population_data

Format

A data frame with 20 rows and 2 variables:

pop

population size

time

time

...

Run drjacoby MCMC

Description

Run MCMC either with or without parallel tempering turned on. Minimum inputs include a data object, a data.frame of parameters, a log-likelihood function and a log-prior function. Produces an object of class drjacoby_output, which contains all MCMC output along with some diagnostics and a record of inputs.

Usage

run_mcmc(
  data,
  df_params,
  misc = list(),
  loglike,
  logprior,
  burnin = 1000,
  samples = 10000,
  rungs = 1,
  chains = 5,
  beta_manual = NULL,
  alpha = 1,
  target_acceptance = 0.44,
  cluster = NULL,
  coupling_on = TRUE,
  pb_markdown = FALSE,
  save_data = TRUE,
  save_hot_draws = FALSE,
  silent = FALSE
)

Arguments

data

a named list or data frame or data values.
df_params

a data.frame of parameters (see ?define_params).
misc

optional list object passed to likelihood and prior. This can be useful for passing values that are not strictly data, for example passing a lookup table to the prior function.
loglike, logprior

the log-likelihood and log-prior functions used in the MCMC. Can either be passed in as R functions (not in quotes), or as character strings naming compiled C++ functions (in quotes).
burnin

the number of burn-in iterations. Automatic tuning of proposal standard deviations is only active during the burn-in period.
samples

the number of sampling iterations.
rungs

the number of temperature rungs used in the parallel tempering method. By default, β\beta values are equally spaced between 0 and 1, i.e. β[i]=\beta[i]=(i-1)/(rungs-1) for i in 1:rungs. The likelihood for the ith heated chain is raised to the power β[i]α\beta[i]^\alpha, meaning we can use the α\alpha parameter to concentrate rungs towards the start or the end of the interval (see the alpha argument).
chains

the number of independent replicates of the MCMC to run. If a cluster object is defined then these chains are run in parallel, otherwise they are run in serial.
beta_manual

vector of manually defined β\beta values used in the parallel tempering approach. If defined, this overrides the spacing defined by rungs. Note that even manually defined β\beta values are raised to the power α\alpha internally, hence you should set alpha = 1 if you want to fix β\beta values exactly.
alpha

the likelihood for the ith heated chain is raised to the power β[i]α\beta[i]^\alpha, meaning we can use the α\alpha parameter to concentrate rungs towards the start or the end of the temperature scale.
target_acceptance

Target acceptance rate. Should be between 0 and 1. Default of 0.44, set as optimum for unvariate proposal distributions.
cluster

option to pass in a cluster environment, allowing chains to be run in parallel (see package "parallel").
coupling_on

whether to implement Metropolis-coupling over temperature rungs. The option of deactivating coupling has been retained for general interest and debugging purposes only. If this parameter is FALSE then parallel tempering will have no impact on MCMC mixing.
pb_markdown

whether to run progress bars in markdown mode, meaning they are only updated when they reach 100% to avoid large amounts of output being printed to markdown files.
save_data

if TRUE (the default) the raw input data is stored for reference in the project output. This allows complete reproducibility from a project, but may be undesirable when datasets are very large.
save_hot_draws

if TRUE the parameter draws relating to the hot chains are also stored inside the pt element of the project output. If FALSE (the default) only log-likelihoods and log-priors are stored from heated chains.
silent

whether to suppress all console output.

Details

Note that both data and misc are passed into log-likelihood/log-prior functions *by reference*. This means if you modify these objects inside the functions then any changes will persist.

Sample posterior draws from all available chains

Description

Sample posterior draws from all available chains

Usage

sample_chains(x, sample_n, keep_chain_index = FALSE)

Arguments

x

an object of class drjacoby_output.
sample_n

An integer number of samples.
keep_chain_index

if TRUE then the column giving the chain is retained.

Value

A data.frame of posterior samples