Usage
tvcm_control(lossfun = neglogLik2,
maxstep = Inf, maxwidth = Inf,
minsize = 30, maxdepth = Inf,
dfpar = 2.0, dfsplit = 0.0,
maxoverstep = ifelse(sctest, Inf, 0),
sctest = FALSE, alpha = 0.05, bonferroni = TRUE,
trim = 0.1, estfun = list(),
maxfacsplit = 5L, maxordsplit = 10, maxnumsplit = 10,
cv = !sctest, folds = folds_control("kfold", 5),
prune = cv, keeploss = FALSE, papply = mclapply,
verbose = FALSE, ...)Arguments
lossfun
a function that extracts a loss measure from a
fitted object, e.g., two times the negative log likelihood
(default).
maxstep
integer. The maximum number of iterations i.e. the
total number of splits processed.
maxwidth
integer (vector). The maximum width of the tree(s).
minsize
numeric. The minimum sum of weights in a terminal
node. The default is the number of varying coefficients times 10.
The parameter specifies also the trimming in parameter coefficient
tests for numeric variables (if sctest = TRUE).
maxdepth
integer (vector). The maximum depth of the tree(s).
dfpar
a numeric scalar larger than zero. The per-parameter
penalty to be applied for stopping. See also argument
maxoverstep.
dfsplit
a numeric scalar larger than zero. The per-split
penalty to be applied for stopping. See also argument
maxoverstep.
maxoverstep
integer scalar. The maximum number of consecutive
times the penalized reduction statistic is allowed to be smaller
than dfsplit before stopping. Specifically, the penalized
loss is computed as the loss (see argument lossfun
sctest
logical scalar. Defines whether coefficient constancy
tests should be used for variable and node selection.
alpha
numeric significance threshold between 0 and 1. A node is
splitted when the smallest (possibly Bonferroni-corrected) $p$
value for any coefficient constancy test in the current step falls
below alpha.
bonferroni
logical. Indicates if and how $p$-values of
coefficient constancy tests must be Bonferroni
corrected. See details.
trim
numeric between 0 and 1. Specifies the trimming parameter
in coefficient constancy tests for continuous partitioning
variables.
maxnumsplit
integer. The maximum number of evaluation for
splits on numeric partitioning variables.
cv
logical scalar. Whether or not the dfsplit parameter
should be cross-validated.
folds
a list of parameters to create folds as produced by
folds_control. prune
logical scalar. Whether or not the overly large
partitions should be pruned by the estimated dfsplit parameter
from cross-validation. Note that prune = TRUE conflicts with
cv = FALSE
keeploss
logical scalar or a numeric equal or larger than
0. Indicates if and how many times the computed penalized loss
reduction statistics should be reused in the following
iterations. Specifically, the option activates approximating the
penaliz
papply
(parallel) apply function, defaults to
mclapply. The function will parallelize the
partition stage and the evaluation of the cross-validation folds as
well as the final pruning stage. verbose
logical. Should information about the fitting process
(such as test statistics, $p$ values,
selected splitting variables and split points) be printed to the
screen?
...
further, undocumented arguments to be passed. These can
include arguments for the papply function.