The hardware and bandwidth for this mirror is donated by dogado GmbH, the Webhosting and Full Service-Cloud Provider. Check out our Wordpress Tutorial.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]dogado.de.
Changes since version 1.1.0.
outliers.enetLTS() lists the observations flagged by
the raw or the reweighted fit with their loss, observed and predicted
response and a type. For the classification families the type separates
leverage points from possible mislabeled observations.assess.enetLTS(), confusion.enetLTS() and
roc.enetLTS() assess a fit on test data.
assess.enetLTS() returns the measures in the layout of
glmnet::assess.glmnet; the deviance is twice the mean
negative log-likelihood.GaussianExample,
BinomialExample and MultinomialExample:
high-dimensional (p > n) data with known outliers and clean test
data. The first two follow the simulation design of Kurnaz, Hoffmann and
Filzmoser (2018).crossfit (gaussian and multinomial): the observations
that are kept are chosen by cross-fitted residuals, so that an
observation is never judged by a model that has seen it, and
alpha and lambda are chosen again on the
reweighted data. The default is NULL, which means
TRUE for family = "multinomial" with
scal = TRUE and FALSE otherwise. The
multinomial fit therefore differs from version 1.1.0 and takes about 2-3
times as long; crossfit = FALSE switches the cross-fitted
trimming off. For the multinomial family the reweighting step flags an
observation when its cross-fitted probability of the observed class is
below a cut-off calibrated to flag a share 2 * del of the
observations if the model is correct. The earlier rule, a median/MAD
bound on the cross-fitted loss within each class, is available with
options(enetLTS.wrule = "mad").screen (multinomial, default TRUE):
observations that are outlying in the predictors are found beforehand,
without the class labels, and left out of the fit. They get weight zero
and are listed in $screened.hsize = "optimal" with hgrid
(experimental): the fraction of retained observations is chosen from
hgrid by cross-validated trimmed loss; the result contains
hsize.search.alpha.raw (gaussian and multinomial): the
mixing parameter of the raw fit. With crossfit = TRUE,
alpha is the value chosen again for the reweighted
fit.plotResid.enetLTS() and
plotDiagnostic.enetLTS() have a new argument
zoom: when the flagged observations are far away, a second
panel shows the regular ones.coef() returns named coefficients
((Intercept), then the column names of xx or
x1, x2, …); for the multinomial family a
matrix with one column per class. With zeros = FALSE it
returns the nonzero coefficients instead of their indices, which are
available from nonzeroCoef.enetLTS().print() shows only the nonzero coefficients, the
selected tuning parameters and the number of flagged observations.$objective is the penalized criterion of the raw fit on
the final subset, on the scale on which the C-steps fit and penalize,
per observation.plotCoef.enetLTS() draws the slopes as needles. The cut-off
lines of the gaussian residual plot are at +/-
qnorm(1 - del), and the residuals are standardized as in
the reweighting step, so the points outside the lines are the flagged
observations. The plots of vers = "reweighted" use the
weights wt. plotResid.enetLTS() and
plotDiagnostic.enetLTS() return the plots invisibly, and
plot.enetLTS() does not change par(ask) or
wait for input.crit.plot is FALSE; the
cross-validation heat map is built without the reshape
package. The packages reshape, grid and
cvTools are no longer imported.predict() accepts a data frame for newX.
cv.enetLTS() accepts class labels for the multinomial
family. enetLTS() stops with an error if xx or
yy contain missing values.glmnet warning
that the smallest lambda values of a path did not converge.weight.binomial() no longer returns NA
weights for observations with a very large linear predictor.residuals(..., vers = "both") for the binomial family
and fitted(..., vers = "both", type = "class") for the
multinomial family return the correct values.mae and raw.mae of the binomial and
multinomial families are computed from the fitted probabilities."a", "b") are coded correctly.alpha with crossfit = TRUE respect
seed and the range of the user-supplied
alphas.plotDiagnostic.enetLTS() no longer sets
options(warn = -1).These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.
Health stats visible at Monitor.