MABC: MICE-assisted Approximate Bayesian Calibration
In wdelva/MABC: Approximate Bayesian Computation with MICE

Description Usage Arguments Details Value

Produces a list with multiple waves of proposed input parameter values to match a vector of target features.

MABC(targets.empirical, model, RMSD.tol.max = 2, min.givetomice = 64,
  n.experiments = 256, start.experiments = NULL, lls, uls,
  strict.positive.params, probability.params, inside_prior = TRUE,
  method = "norm", predictorMatrix = "complete", maxit = 50,
  maxwaves = 4, n_cores = n_cores, multinode = FALSE)

`targets.empirical`	The vector of target features
`model`	Wrapper function for the simulation model. See details for a description of the required format.
`RMSD.tol.max`	Tolerance for the root mean squared distance between target features and model output features
`min.givetomice`	Minimal number of observations in the training dataset to which MICE is applied
`n.experiments`	Number of parameter combinations in each wave of model runs
`start.experiments`	If set to NULL (default), start experiments will be drawn uniformly from the prior distributions. If a matrix of input parameter values, possibly output from a previous calibration, models with these inputs will be run, instead of drawing from the prior distributions. To resume where a previous calibration ended, you can input start.experiments as a data.frame with inputs, outputs, seed, wave and RMSD values.
`lls`	Vector of lower limits of the prior distribution of input parameter values
`uls`	Vector of upper limits of the prior distribution of input parameter values
`strict.positive.params`	Vector of indices that indicate which of the input parameters are strictly positive. Set to zero if there are no such parameters.
`probability.params`	Vector of indices that indicate which of the input parameters are strictly between 0 and 1. Set to zero if there are no such parameters.
`inside_prior`	TRUE by default. If FALSE, parameter sampling is not restricted to the initial ranges of the prior distribution during the subsequent algorithm steps.
`method`	Method used by MICE. Currently, only "norm" is supported.
`predictorMatrix`	As in mice::mice. Can be "complete", "LASSO" (Least Absolute Shrinkage and Selection Operator), or a user-defined matrix of zeros and ones. Diagonal must always be zeros. Ones indicate which variables are included in the chained equations in MICE.
`maxit`	The maxit argument used in MICE (number of times that the chained equations are cycled through)
`maxwaves`	The maximum number of waves of model runs
`n_cores`	The number of cores available for parallel model runs. Default = 1, i.e. serial execution of model runs
`multinode`	TRUE or FALSE (Default). If TRUE, model runs are distributed over the cores of multiple nodes, using DOsnow and snow as the back-end to the foreach package. If FALSE and n_cores > 1, model runs are distributed over the cores of a single node, using the parallel package.

The model wrapper function for the simulaton model must have a vector of model input parameter values as its one and only argument. Furthermore, it must return a vector of model features. These model features are then compared against the target features.

A list with the following components:

`results`	A list with `maxwaves` elements. Each element is itself a list with 6 named elements: `thiswave.io`: Input and Output of the simulations of this wave, `thiswave.median.features`: The l1 median of this wave's model output, `upto.thiswave.median.features`: The l1 median of all model output up to and including this wave, `closest.upto.thiswave.median.features`: The l1 median of the `min.givetomice` model features that are closest to the target features, up to this wave, `max.RMSD`: The largest Root Mean Square Distance of the `min.givetomice` model features that are closest to the target features, up to this wave. The `max.RMSD` is expected to decrease with increasing waves, `closest.upto.thiswave.io`: Input and Output of the `min.givetomice` simulations that have features closest to the target features, up to this wave. The input parameters make up the approximate posterior distribution.
`targets`	The vector of target features
`secondspassed`	The time (in seconds) it took MABC to complete the calibration