singleStep: Updating mechanisms used in the metropolis algorithm
In ClementCalenge/metroponcfs: Metropolis algorithm for model fitting

Description Usage Arguments Details Value Author(s) See Also

Updating mechanisms used in the metropolis algorithm

singleStep(par, lidat, nam, pum, ctrl, logposterior, postcur, optimiser, debug,
  dumpfile, composition)

MultipleIndependentSteps(par, lidat, nam, pum, ctrl, logposterior, postcur,
  optimiser, debug, dumpfile, composition)

MultiNormalStep(par, lidat, nam, pum, ctrl, logposterior, postcur, optimiser,
  debug, dumpfile, composition)

`par`	a named list containing the parameters of the model (the names of the list correspond to the names of the parameters)
`lidat`	a named list containing the variables and constants required by the model (especially the function `logposterior`
`nam`	character string. The name of the parameter to be updated.
`pum`	for `singleStep`, the standard deviation of the Gaussian distribution used to generate the proposal. For `MultipleIndependentSteps`, a vector containing the standard deviations for the Gaussian distribution used to generate the proposal of every element of the vector. For `MultiNormalStep`, the covariance matrix of the multinormal distribution.
`ctrl`	a named list with two parameters: (i) `namePar` is the name of the parameter in `par` that is updated, (ii) if this parameter is a vector where each component is updated in turn (e.g. for `MultipleIndependentSteps`), `iter` indicates the element of this vector that is updated (this element is ignored in other cases).
`logposterior`	a function used to calculate the log-posterior probability of `par` (up to a constant). This function must have two arguments: `par` is the list of parameters and `lidat` is the list containing the data used in the model.
`postcur`	the current value of the posterior, before updating.
`optimiser`	logical. If `TRUE`, the log-posterior calculation is optimized for the current parameter (i.e. the posterior is proportional to the log-posterior but the proportionality constant is not necessarily the same for other optimized parameters, so that the current posterior should be calculated before every step). If `FALSE` the function `logposterior` returns the exact log-posterior. In other words, if `optimiser` is TRUE, the function `singleStep`, etc. calculates the current posterior (prior to updating) and does not take into account the value of `postcur` passed to the function. If FALSE, the function relies on the value of `postcur` passed as argument for the updating.
`debug`	logical. If `TRUE` the updating should be run in debugging mode.
`dumpfile`	character. The name of the debugging file.
`composition`	character. How composition of parameters are managed (see the help page of `GeneralSingleMetropolis`). Currently, only used by `MultipleIndependentSteps`.

singleStep implements an updating mechanism for a unique parameter based on a normal distribution. MultipleIndependentSteps implements the same updating mechanism for a vector of parameters. MultiNormalStep implements an updating mechanism relying on the sampling of a multinormal distribution. These functions are used by the package, but are not to be used directly by the user.

A list with three elements: (i) parm is the updated vector of parameters, (ii) postcur is the value of the (optionally optimized) posterior after updating, and (iii) accept is a vector containing TRUE if an element proposed has been accepted and FALSE otherwise

Clement Calenge, clement.calenge@oncfs.gouv.fr

GeneralSingleMetropolis

ClementCalenge/metroponcfs documentation built on May 6, 2019, 12:05 p.m.