This is an internal function of package mgcv which allows
control of the numerical options for fitting a GAM.
Typically users will want to modify the defaults if model fitting fails to
converge, or if the warnings are generated which suggest a
loss of numerical stability during fitting. To change the default
choise of fitting method, see gam arguments method and optimizer.
gam.control(nthreads=1,irls.reg=0.0,epsilon = 1e-07, maxit = 200,
mgcv.tol=1e-7,mgcv.half=15, trace = FALSE,
rank.tol=.Machine$double.eps^0.5,nlm=list(),
optim=list(),newton=list(),outerPIsteps=0,
idLinksBases=TRUE,scalePenalty=TRUE,efs.lspmax=15,
efs.tol=.1,keepData=FALSE,scale.est="fletcher",
edge.correct=FALSE)Some parts of some smoothing parameter selection methods (e.g. REML) can use some
parallelization in the C code if your R installation supports openMP, and nthreads is set to
more than 1. Note that it is usually better to use the number of physical cores here, rather than the
number of hyper-threading cores.
For most models this should be 0. The iteratively re-weighted least squares method
by which GAMs are fitted can fail to converge in some circumstances. For example, data with many zeroes can cause
problems in a model with a log link, because a mean of zero corresponds to an infinite range of linear predictor
values. Such convergence problems are caused by a fundamental lack of identifiability, but do not show up as
lack of identifiability in the penalized linear model problems that have to be solved at each stage of iteration.
In such circumstances it is possible to apply a ridge regression penalty to the model to impose identifiability, and
irls.reg is the size of the penalty.
Maximum number of IRLS iterations to perform.
The convergence tolerance parameter to use in GCV/UBRE optimization.
If a step of the GCV/UBRE optimization method leads to a worse GCV/UBRE score, then the step length is halved. This is the number of halvings to try before giving up.
Set this to TRUE to turn on diagnostic output.
The tolerance used to estimate the rank of the fitting problem.
list of control parameters to pass to nlm if this is
used for outer estimation of smoothing parameters (not default). See details.
list of control parameters to pass to optim if this
is used for outer estimation of smoothing parameters (not default). See details.
list of control parameters to pass to default Newton optimizer used for outer estimation of log smoothing parameters. See details.
The number of performance interation steps used to initialize outer iteration.
If smooth terms have their smoothing parameters linked via
the id mechanism (see s), should they also have the same
bases. Set this to FALSE only if you are sure you know what you are doing
(you should almost surely set scalePenalty to FALSE as well in this
case).
gamm is somewhat sensitive to the absolute scaling
of the penalty matrices of a smooth relative to its model matrix. This option rescales
the penalty matrices to accomodate this problem. Probably should be set to FALSE
if you are linking smoothing parameters but have set idLinkBases to FALSE.
maximum log smoothing parameters to allow under extended Fellner Schall smoothing parameter optimization.
change in REML to count as negligible when testing for EFS convergence. If the step is small and the last 3 steps led to a REML change smaller than this, then stop.
Should a copy of the original data argument be kept in the gam
object? Strict compatibility with class glm would keep it, but it wastes space to
do so.
How to estimate the scale parameter for exponential family models estimated
by outer iteration. See gam.scale.
With RE/ML smoothing parameter selection in gam using the default Newton RE/ML optimizer, it is possible to improve inference at the
`completely smooth' edge of the smoothing parameter space, by decreasing
smoothing parameters until there is a small increase in the negative RE/ML (e.g. 0.02). Set to TRUE or to a number representing the target increase to use. Only changes the corrected smoothing parameter matrix, Vc.
Outer iteration using newton is controlled by the list newton
with the following elements: conv.tol (default
1e-6) is the relative convergence tolerance; maxNstep is the maximum
length allowed for an element of the Newton search direction (default 5);
maxSstep is the maximum length allowed for an element of the steepest
descent direction (only used if Newton fails - default 2); maxHalf is
the maximum number of step halvings to permit before giving up (default 30).
If outer iteration using nlm is used for fitting, then the control list
nlm stores control arguments for calls to routine
nlm. The list has the following named elements: (i) ndigit is
the number of significant digits in the GCV/UBRE score - by default this is
worked out from epsilon; (ii) gradtol is the tolerance used to
judge convergence of the gradient of the GCV/UBRE score to zero - by default
set to 10*epsilon; (iii) stepmax is the maximum allowable log
smoothing parameter step - defaults to 2; (iv) steptol is the minimum
allowable step length - defaults to 1e-4; (v) iterlim is the maximum
number of optimization steps allowed - defaults to 200; (vi)
check.analyticals indicates whether the built in exact derivative
calculations should be checked numerically - defaults to FALSE. Any of
these which are not supplied and named in the list are set to their default
values.
Outer iteration using optim is controlled using list
optim, which currently has one element: factr which takes
default value 1e7.
Wood, S.N. (2011) Fast stable restricted maximum likelihood and marginal likelihood estimation of semiparametric generalized linear models. Journal of the Royal Statistical Society (B) 73(1):3-36
Wood, S.N. (2004) Stable and efficient multiple smoothing parameter estimation for generalized additive models. J. Amer. Statist. Ass.99:673-686.