R-package FME contains functions to run complex applications of models
that produce output as a function of input parameters.
Although it was created to be used with models consisting of ordinary
differential equations (ODE), partial differential equations (PDE) or
differential algebraic equations (DAE), it can work with other models.
Functions to allow fitting of the model to data.
Function modCost
estimates the (weighted) residuals between
model output and data, variable and model costs.
Function modFit
uses the output of modCost
to find
the best-fit parameters. It provides a wrapper around R's
built-in minimisation routines (optim
, nlm
,
nlminb
) and nls.lm
from package minpack.lm
.
Package FME also includes an implementation of the pseudo-random
search algorithm (function pseudoOptim
).
Function sensFun
estimates the sensitivity functions of
selected output variables as a function of model parameters. This
is the basis of uni-variate, bi-variate and multi-variate
sensitivity analysis.
Function collin
uses as input the sensitivity functions
and estimates the "collinearity" index for all possible parameter
sets. This multivariate sensitivity estimate measures approximate
linear dependence and is useful to derive which parameter sets are
identifiable given the data set.
Function sensRange
produces 'envelopes' around the
sensitivity variables, consisting of a time series or a
1-dimensional set, as a function of the sensitivity parameters.
It produces "envelopes" around the variables.
Function modCRL
calculates the values of single
variables as a function of the sensitivity parameters. This
function can be used to run simple "what-if" scenarios
Function modMCMC
runs a Markov chain Monte Carlo
(Bayesian analysis). It implements the delayed rejection -
adaptive Metropolis (DRAM) algorithm.
FME also contains functions to generate multiple parameter
values arranged according to a grid (Grid
) multinormal
(Norm
) or uniform (Unif
) design, and a latin
hypercube sampling (Latinhyper
) function