The new smof argument opt.method is introduced
from which a wider choice of numerical optimizers is made available,
that is, three optimizers from the base package are accessibile.
In addition, if the package nloptr is installed on the local
system, various additional algorithms are made available.
When the transformations are computed using quantiles of distribution,
the final values of the scores are now mapped to the interval
(1,K), if K is the number of levels of a given facrtor.
The reasons of this choice are explained in the ‘Details’
section of the smof documentation page.
Generally improved and more detailed documentation.
Objects of class gnm are now handled.
Methods coef and anova are introduced for class smof.
The smof argument trace is replaced by the more flexible
argument verbose.
New smof argument f.tail allows to select the
suffix being appended to the names the factors components.
Fixed a bug in the parsing of the object formula which caused problems when some factor has a name which is portion of another factor name.
Fixed a bug in smof_refit affecting the retrival of the
work parameters estimated by a previous call to smof.
A variant method for assigning numeric scores is introduced, based on a monotonic spline instead of the quantile function of a distribution.
An implication of the new variant method is that smof argument
scoring had to redesigned.
The new function smof_refit is introduced.
Additional arguments of smof are introduced:
original and opt.control.
Fixed a bug in predict.smof which affected the case with multiple factors.
The object returned by smof now includes additional components
and is assigned to class smof.
Various methods are introduced for class smof:
print, summary, plot, predict.
Fixed a bug which prevented the use of option fast.fit=TRUE
when family=binomial(..) and the response was of binary type.
First public version of the package and upload to CRAN, with
only one exported function, namely smof.
Scoring of factors are produced from suitably chosen quantiles of
distributions among the families "gh", "SU", "sinh-arcsinh"
or their equivalente names.
Admissible classes of fitted models are "lm", "mlm", "glm",
"survreg", "coxph.penal", "coxph".