Control Hyper-parameters for Boosting Algorithms
Definition of the initial number of boosting iterations, step size and other hyper-parameters for boosting algorithms.
boost_control(mstop = 100, nu = 0.1, risk = c("inbag", "oobag", "none"), stopintern = FALSE, center = TRUE, trace = FALSE)
an integer giving the number of initial boosting iterations. If
mstop = 0, the offset model is returned.
a double (between 0 and 1) defining the step size or shrinkage parameter. The default is probably too large for many applications with
family = Poisson()and a smaller value is better.
a character indicating how the empirical risk should be computed for each boosting iteration.
inbagleads to risks computed for the learning sample (i.e., all non-zero weights),
oobagto risks based on the out-of-bag (all observations with zero weights) and
noneto no risk computations at all.
a logical that defines if the boosting algorithm stops internally when the out-of-bag risk in one iteration is larger than the out-of-bag risk in the iteration before. Can also be a positive number giving the risk difference that needs to be exceeded.
deprecated. A logical indicating if the numerical covariates should be mean centered before fitting. Only implemented for
blackboostcentering is not needed. In
gamboostcentering is only needed if
bolsbase-learners are specified without intercept. In this case centering of the covariates is essential and should be done manually (at the moment). Will be removed in favour of a corresponding argument in
glmboostin the future (and gives a warning).
a logical triggering printout of status information during the fitting process.
An object of class
boost_control, a list.