Inversion with automatic convergence checking

Inversion with automatic convergence checking

invert.auto(
  observed,
  invert.options,
  return.samples = TRUE,
  save.samples = NULL,
  quiet = FALSE,
  parallel = TRUE,
  parallel.cores = NULL,
  parallel.output = "/dev/null"
)

Arguments

observed

Vector, matrix, or data frame (coerced to matrix) of observed values. For spectral data, wavelengths are rows and spectra are columns. Dimensions must align with the output of model.

invert.options

Parameters related to inversion.

return.samples

Include full samples list in output. Default = TRUE.

save.samples

Save samples to file as the inversion proceeds (useful for debugging). If NULL, do not save samples. Default = NULL.

quiet

Suppress progress bar and status messages. Default=FALSE

parallel

Logical. Whether or not to run multiple chains in parallel on multiple cores (default = TRUE).

parallel.cores

Number of cores to use for parallelization. If NULL (default), allocate one fewer than detected number of cores.

parallel.output

Filename (or ” for stdout) for printing parallel outputs. Use with caution. Default = '/dev/null'.

Value

List including results (summary statistics), samples (mcmc.list object, or NA if return.samples=FALSE), and other information.

Details

Performs an inversion via the invert.custom function with multiple chains and automatic convergence checking. Convergence checks are performed using the multivariate Gelman-Rubin diagnostic.

Parameters specific to invert.auto are described here. For the remaining parameters, see invert.custom().

  • model – The model to be inverted. This should be an R function that takes params as input and returns one column of observed (nrows should be the same). Constants should be implicitly included here.

  • nchains – Number of independent chains.

  • inits.function – Function for generating initial conditions.

  • ngibbs.max – Maximum number of total iterations (per chain). DEFAULT = 5e6

  • ngibbs.min – Minimum number of total iterations (per chain). DEFAULT = 5000.

  • ngibbs.step – Number of iterations between convergence checks. Default = 1000.

  • run_first – Function to run before running sampling. Takes parallel inputs list containing runID, initial values, and resume (NULL) as an argument.

  • calculate.burnin – If TRUE, use PEcAn.assim.batch::autoburnin function to calculate burnin. Otherwise, assume burnin is min(niter/2, iter_conv_check).

  • threshold – Maximum value of the Gelman-Rubin diagnostic for determining convergence. Default = 1.1