o Experimental: multivariate ridge regression. These
functions are new in this version; their arguments and results may
change in future releases. - ridge() now accepts a matrix
response: cbind(Y1, Y2) ~ ... in a formula, a multivariate
lm() fit, or a matrix y. All responses are
fitted with common ridge constants, and the result has class
"mlmridge". The implementation was adapted from
lm_multiv_ridge() in the VARshrink
package. - The HKB, LW and GCV criteria are given for each response; GCV
can also be pooled over responses (pool = "scaled",
"raw" or "mahalanobis") to choose a common
ridge constant. - as.ridge() extracts the results for one
response as an ordinary "ridge" object. -
coef(), vcov() and print()
methods for "mlmridge" objects. - best()
method for "mlmridge" objects: the optimal ridge constants
for each response and criterion, plus GCV pooled over responses, as a
data frame; best(x, response = ) gives one response. -
traceplot(), plot(), pairs(),
plot3d(), biplot(), pca(),
precision() and vif() methods apply the
univariate methods to one response at a time (response =).
traceplot() by default draws all responses, one panel each,
marking the ridge constant chosen by pooled GCV
(GCV = TRUE). - plot(x, predictor = ) draws
coefficient ellipses for one or more predictors in the space of two
responses (responses =), traced over the ridge constants. -
A one-column matrix response (e.g., cbind(y) ~ ...) gives
an ordinary "ridge" result.
o New vignette, “Getting started with genridge”, introducing the main
functions with the longley and Detroit
examples. o traceplot() is now an S3 generic, with methods
for "ridge" and "mlmridge" objects. o
traceplot() and the plot() method for
vif() results now leave enough room for the variable
labels, based on their width, so long names are no longer clipped at the
side of the plot.
o New ridge.lm() method: ridge() can now be
applied directly to a model fitted by lm(), e.g.,
ridge(lm(y ~ x1 + x2, data), lambda = ...). The response,
model matrix (with any factor contrasts and transformed terms) and
offset are taken from the fitted model. glm and weighted
lm fits give an error; a multivariate lm() fit
gives a multivariate ridge regression (see above).
o Documentation converted to roxygen markdown
(Roxygen: list(markdown = TRUE)): \code{},
\link{}, \emph{} and \bold{}
replaced by markdown equivalents, and \describe{},
\itemize{}, \enumerate{} and
@return \item{} lists by markdown lists. Also
fixed a few small errors in the docs.
o ridge() with a formula and svd = FALSE
failed with “variable lengths differ”. o ridge() with a
formula failed when the model had only one predictor; vif()
failed for such a model. o The HKB and LW criteria, which have the
factor (p - 2), are now NA for fewer than three predictors
(they were 0 for p = 2 and negative for p = 1), and the plot methods no
longer mark them. o best() was documented but its generic
was not exported; it now is. o best() reported the same
(wrong) df for all three criteria; each now gets its own
equivalent degrees of freedom.
o Replaced links to datavis.ca, which is offline, with
DOIs (documentation) and friendly.github.io/papers
(README).
o Added a testthat test suite: the multivariate fit against the
univariate fits and the closed-form estimator, its covariance structure,
the pooled GCV criteria, best(), the methods for
"mlmridge" objects, and the univariate results against
saved reference values from 0.8.1.
o genridge now requires R >= 4.1.0 (for the native pipe,
|>).
This is a major update to the package adding additional plotting methods
o Fleshed out and installed plot.precision() o
Illustrate plot.precision() in README.Rmd o
Added diab data, diabetes from CASI o Fixed documentation
for plot.ridge() so that plot.pcaridge() is
documented explicitly. o vif.ridge() now returns a
"vif.ridge" object list to prepare for a plot method. o
Added plot.vif.ridge() to plot VIFs vs shrinkage o Added
norm.diff = sqrt((b_OLS - b_ridge)^2) as a measure of
shrinkage to precision() o Added
plot.precision() for plots of shrinkage vs. precision o
Added best() to display the optimal shrinkage criteria
o Added links to gentalk.pdf o Improved README and fixed some
examples o precision() result gains a class “precision” in
preparation for a plot method o Implemented
plot.precision() for plots shrinkage vs. variance, using
various criteria o Fixed warning from ridge() related to
contrasts o ridge() now collects the optimal
criteria in a named list, which can be added in
plot.precision()
o Converted the package to roxygen2 documentation, correcting some infelicities with S3 methods o Added an extended README example. o fix link to genridge paper PDF
o a maintenance release, correcting a problem with an in one .Rd file
o Remove references to ElemStatLearn, now defunct
o bump package version, preparing to move to Github
o Use rgl:: in calls to rgl package for CRAN o Fixed long lines in .Rd files
o Added contourf()
o Minor tweaks to plot3d.ridge o Added biplot.ridge method for symmetry, showing the PCA vectors in variable space o Added examples to ridge.Rd using data(manpower, package=“bestglm”) o Added Manpower data
o Made ridge() an S3 generic with formula and default methods o Fixed labels problems in biplot.pcaridge and plot3d.ridge o Added Detroit data with examples showing most methods
o Fixed order of classes in pca.ridge o Made plot3d an S3 generic; added plot3d.pcaridge and plot.pcaridge o Added biplot.pcaridge() to draw variable vectors on a plot.pcaridge() plot o Added options to precision.* for greater flexibility o Added GCV computation to ridge() o Added which.lambda to select ellipses in plot.ridge(), plot3d.ridge() etc. o Added labels argument to plot.ridge() and fixed pairs.ridge for this.
o Added plot3d.ridge() using rgl, completing the main stages in visualization methods for ridge-like problems
o Added precision.ridge() method for measures of precision and shrinkage o Added vif.ridge() method, extending the car S3 generic for ridge objects o Now Depends: car
o Added Acetylene data o Added vcov.ridge() method
o Fixed Authors@R in DESCRIPTION
o Added NAMESPACE o Initial package version, released to R-Forge o Added pca.ridge