MINI MINI MANI MO

%
% Copyright (c) 2014, 2016, Oracle and/or its affiliates. All rights reserved.
%
\name{ore.glm}
\alias{ore.glm}
\alias{ore.glm.control}
\alias{summary.ore.glm}
\alias{print.summary.ore.glm}
\alias{predict.ore.glm}
\alias{residuals.ore.glm}
\alias{vcov.ore.glm}
\concept{regression}
\title{Oracle R Enterprise Generalized Linear Models}
\description{
  Functions for fitting and using generalized linear models on
  \code{ore.frame} data.
}
\usage{
### Fitting function
ore.glm(formula, data, weights, family = gaussian(), start = NULL,
        control = list(...), contrasts = NULL, xlev = NULL,
        ylev = NULL, yprob = NULL, ...)

### Fit control function
ore.glm.control(devlre = 8, maxit = 25, linesearch = FALSE,
                trace = getOption("ore.trace", FALSE), ...)

### Specific methods for ore.glm objects
\S3method{summary}{ore.glm}(object, dispersion = NULL, correlation = FALSE,
        symbolic.cor = FALSE, ...)

\S3method{vcov}{ore.glm}(object, ...)

\S3method{predict}{ore.glm}(object, newdata = NULL, type = c("link", "response"),
        se.fit = FALSE, dispersion = NULL, na.action = na.pass, supplemental.cols = NULL,
        ...)

\S3method{residuals}{ore.glm}(object, type = c("deviance", "pearson", "working", "response", "partial"),
          ...)

### Inherited methods for ore.glm objects
#coef(object, ...)
#coefficients(object, ...)
#deviance(object, ...)
#effects(object, ...)
#extractAIC(fit, scale, k = 2, ...)
#family(object, ...)
#fitted(object, ...)
#fitted.values(object, ...)
#formula(x, ...)
#logLik(object, ...)
#model.frame(formula, ...)
#nobs(object, ...)
#weights(object, ...)
}
\arguments{
  \item{formula}{A \code{\link[stats]{formula}} object representing the
    model to be fit.}
  \item{data}{An \code{ore.frame} object specifying the data for the
    model.}
  \item{weights}{An optional \code{ore.number} object specifying the
    analytic weights in the model.}
  \item{ supplemental.cols }{ Additional columns to include in the prediction
  result from the \code{newdata} data set. }
  \item{family}{A \code{\link[stats]{family}} object specifying the
    generalized linear model family details. This is the same type of
    family object that is used by the \code{\link[stats]{glm}} function
    in the \pkg{stats} package, which includes the
    \code{\link[MASS]{negative.binomial}} function from the \pkg{MASS}
    package as well as the \code{\link[statmod]{tweedie}} function from
    the \pkg{statmod} package.}
  \item{start}{An optional \code{\link[base]{numeric}} vector specifying
    the initial coefficient estimates in the linear predictor.}
  \item{control}{An optional \code{\link[base]{list}} object containing
    a list of fit control parameters to be interpreted by the
    \code{ore.glm.control} function.}
  \item{contrasts}{An optional named \code{\link[base]{list}} to be
    supplied to the \code{contrasts.arg} argument of
    \code{\link[stats]{model.matrix}}.}
  \item{xlev}{An optional named \code{\link[base]{list}} of
    \code{\link[base]{character}} vectors specifying the
    \code{\link[base]{levels}} for each
    \code{\link[OREbase:ore.factor-class]{ore.factor}} variable.}
  \item{ylev}{An optional \code{\link[base]{character}} vector to
    specify the response variable levels in
    \code{\link[stats]{binomial}} generalized linear models.}
  \item{yprob}{An optional numeric value between 0 and 1 specifying the
    overall probability of \code{y != ylev[1]} in
    \code{\link[stats]{binomial}} generalized linear models.}
  \item{devlre}{A positive number specifying the minimum log relative
    error of the residual deviance convergence criterion,
    \eqn{-log10(|dev - dev_{old}|/|dev|) \ge devlre}.}
  \item{maxit}{A positive integer specifying the maximum number of
    Fisher scoring iterations.}
  \item{linesearch}{A logical value indicating whether a line search
    should be used after each Fisher scoring iteration. If \code{FALSE},
    then a line search will used after the first two iterations and any
    subsequent iteration that results in an increase in the residual
    deviance.}
  \item{trace}{The control parameter that controls the output produced
    at each Fisher scoring iteration;
    a value of \code{FALSE} or \code{0} indicating no output,
    a value of \code{TRUE} or \code{1} indicating the printing of the
    residual deviance for each iteration, or
    a value of \code{2} indicating the printing of the residual deviance
    and runtime breakdown for each iteration. The default value is
    regulated by the global option \code{ore.trace}.}
  \item{object, newdata}{An \code{ore.glm} object.}
  \item{dispersion, correlation, symbolic.cor}{Argument not implemented.}
  \item{type}{A character string specifying the type of predictions or
    residuals to produce.}
  \item{se.fit}{A logical value indicating whether to return the
    standard errors for the predictions.}
  \item{na.action}{The manner in which \code{NA} values are handled,
    either \code{na.omit} or \code{na.pass}.}
  \item{\dots}{Additional arguments.}
}
\details{
  The \code{ore.glm} function fits generalized linear models using a
  Fisher scoring iteratively re-weighted least squares (IRLS)
  algorithm. Instead of the traditional step halving to prevent the
  selection of less optimal coefficient estimates, a line search is used
  to select new coefficient estimates at each iteration starting from
  the current coefficient estimates and moving through the Fisher
  scoring suggested estimates using the formula
  \eqn{(1 - \alpha) * old + \alpha * suggested} where \eqn{\alpha} in
  \eqn{[0, 2]}.

  Each iteration consists of up to three embedded R map/reduce
  operations: an IRLS operation and up to two line search operations.
  The IRLS map/reduce operations are on the matrix cross-products based
  off of \code{model.matrix} or \code{sparse.model.matrix} function
  calls depending on the underlying sparsity of the model matrix. After
  the algorithm has either converged or reached the maximum number of
  iterations, a final embedded R map/reduce operation is used to
  generate the complete set of model-level statistics.

  The \code{\link[OREbase:ore.options]{"ore.parallel"}} global option is
  used by \code{ore.glm} to determine the preferred degree of
  parallelism to use within the Oracle R Enterprise server.
}
\value{
  For \code{ore.glm}, returns an \code{ore.glm} object.

  For \code{summary.ore.glm}, returns a \code{summary.ore.glm} object.

  Note: the training data referenced by argument \code{data} is needed to
  produce meta information about the \code{ore.glm} object.
}
\references{
  \href{http://www.oracle.com/technetwork/database/database-technologies/r/r-enterprise/documentation/index.html}{Oracle R Enterprise}
}
\author{
  Oracle \email{oracle-r-enterprise@oracle.com}
}
\seealso{
  \link[OREstats]{model.matrix,formula-method} (\pkg{OREstats} package),
  \code{\link{ore.lm}},
  \code{\link[stats]{glm}},
  \code{\link[stats]{family}},
  \code{\link[OREbase:ore.options]{ore.parallel}}
}
\examples{
\dontshow{
if (!interactive())
    ore.connect(user     = Sys.getenv("ORE_USERNAME", "rquser"),
                sid      = Sys.getenv("ORACLE_SID"),
                host     = Sys.getenv("HOST"),
                password = Sys.getenv("ORE_PASSWORD", "rquser"),
                port  = if (.Platform$OS.type == "windows")
                           Sys.getenv("ORACLE_PORT")
                        else
                           Sys.getenv("TCPPORT"),
                all = TRUE)
}
  # Load libraries for examples
  library(OREstats)
  library(rpart)   # kyphosis and solder data sets

  # Logistic regression
  KYPHOSIS <- ore.push(kyphosis)
  kyphFit1 <- ore.glm(Kyphosis ~ ., data = KYPHOSIS, family = binomial())
  kyphFit2 <- glm(Kyphosis ~ ., data = kyphosis, family = binomial())
  summary(kyphFit1)
  summary(kyphFit2)

  # Poisson regression
  SOLDER <- ore.push(solder)
  solFit1 <- ore.glm(skips ~ ., data = SOLDER, family = poisson())
  solFit2 <- glm(skips ~ ., data = solder, family = poisson())
  summary(solFit1)
  summary(solFit2)

  # Negative binomial regression
  solFit3 <- ore.glm(skips ~ ., data = SOLDER, family = MASS::negative.binomial(10))
  solFit4 <- glm(skips ~ ., data = solder, family = MASS::negative.binomial(10))
  print(summary(solFit3))
  print(summary(solFit4))

  # Tweedie regression
  solFit5 <- ore.glm(skips ~ ., data = SOLDER, family = statmod::tweedie(1.5))
  solFit6 <- glm(skips ~ ., data = solder, family = statmod::tweedie(1.5))
  print(summary(solFit5))
  print(summary(solFit6))
}
\keyword{regression}