configureHMC {nimbleHMC}

Configure HMC

Description

Create a nimble MCMC configuration object which applies HMC sampling to continuous-valued dimensions

Usage

configureHMC(
  model,
  nodes = character(),
  type = "NUTS",
  control = list(),
  print = TRUE,
  ...
)

Arguments

model

A nimble model, as returned by 'nimbleModel'

nodes

A character vector of stochastic node names to be sampled. If an empty character vector is provided (the default), then all stochastic non-data nodes will be sampled. An HMC sampler will be applied to all continuous-valued non-data nodes, and nimble's default sampler will be assigned for all discrete-valued samplin to apply, either "NUTS" or "NUTS_classic".

type

A character string specifying the type of HMC sampling to apply, either "NUTS" or "NUTS_classic". See 'help(NUTS)' or 'help(NUTS_classic)' for details of each sampler. The default sampler type is "NUTS".

control

Optional named list of control parameters to be passed as the 'control' argument to the HMC sampler. See 'help(NUTS)' or 'help(NUTS_classic)' for details of the control list elements accepted by each sampler.

print

Logical argument specifying whether to print the montiors and samplers. Default is TRUE.

...

Other arguments that will be passed to 'configureMCMC'

Details

This function can be used like ‘configureMCMC' in nimble to create an MCMC configuration object. It will return an MCMC configuration with an HMC sampler assigned to continuous-valued model dimensions, and nimble’s default sampler assigned for discrete-valued dimensions (or, only for the nodes specified in the 'nodes' argument). The resulting MCMC configuration object can be used as an argument to 'buildMCMC' to generate an executable MCMC algorithm.

Either the 'NUTS_classic' or the 'NUTS' sampler can be applied. Both implement variants of No-U-Turn HMC sampling, however the 'NUTS' sampler uses more modern adapatation techniques. See 'help(NUTS)' or 'help(NUTS_classic)' for details.

Use this function if you want to create an MCMC configuration, and then modify it further before building the MCMC algorithm. 'buildHMC' provides a more direct route to a compilable MCMC algorithm with HMC sampling applied to all continuous-valued dimensions.

Value

An object of class 'MCMCconf'.

Author(s)

Daniel Turek

See Also

addHMC buildHMC configureMCMC addSampler sampler_NUTS sampler_NUTS_classic

Examples

code <- nimbleCode({
    b0 ~ dnorm(0, 0.001)
    b1 ~ dnorm(0, 0.001)
    sigma ~ dunif(0, 10000)
    for(i in 1:N) {
        mu[i] <- b0 + b1 * x[i]
        y[i] ~ dnorm(mu[i], sd = sigma)
    }
})

N <- 10
constants <- list(N = N, x = 1:N)
data <- list(y = 1:N)
inits <- list(b0 = 1, b1 = 0.1, sigma = 1)

Rmodel <- nimbleModel(code, constants, data, inits, buildDerivs = TRUE)

## create MCMC configuration object with only an HMC sampler
conf <- configureHMC(Rmodel)

Rmcmc <- buildMCMC(conf)

# Cmodel <- compileNimble(Rmodel)
# Cmcmc <- compileNimble(Rmcmc, project = Rmodel)
# samples <- runMCMC(Cmcmc)