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.
cv.varpro() gains optional outer cross-validation
through cv.folds and foldid, with held-out
prediction errors, selection frequencies, and importance-stability
summaries for all three selection rules. The default,
cv.folds = 0, retains OOB-based selection without outer
cross-validation.varpro() objects now include model.info,
recording original and working responses, survival targets and RMST
horizons, class-label mappings, and effective forest settings.
model.info$observations also records input, retained, and
omitted row counts, including in split.weight.only results
(#7).plot() for iVarPro objects no longer supports
ladder, ladder.cuts, or
ladder.max.segments; remove these arguments from existing
calls.varpro(), partialpro(),
rf.learner(), gbm.learner(), and
bart.learner() now reject unnamed options, unrecognized
names, and duplicate names in ... (#7). This includes
unsupported na.action in varpro() and
RMST, rmst, and time in
partialpro().rf.learner() now rejects user-supplied
formula, data, xvar.wt, and
perf.type, which the wrapper sets internally (#7).get.orgvimp() now
reuses a supplied vmp summary.ivarpro() to honor scale = "global"
and scale = "none" throughout neighborhood searches and
exclude invalid local fits with use.loo = FALSE.ivarpro() and preserved all columns of user-supplied
y.external responses.noise.na overrides, and restore
training OOB scores when predict() is called without
newdata, including on a prediction result.nblocks = 1 now respects supplied held-out
data.cv.varpro() input validation, cutoff ordering,
and handling of unavailable prediction errors, with warnings when no
candidate error is finite.get.rmst() to use full-ensemble survival
estimates with an OOB fallback, validate time horizons, and preserve
matrix dimensions for small inputs and response identities for multiple
horizons. The integral now uses survival at each interval’s left
endpoint (#7).varpro.strength(..., stat = "oob") to return
meanOOB for regression.sdependent() to align importance-matrix rows and
columns by variable name before clearing self-links, fill missing
release rows with zeros, and validate matrix values and names.varpro() now warns when preprocessing omits
observations with missing values, reporting the input, omitted, and
retained counts (#7).shap.ivarpro() with the iVarPro plotting
methods and get.beta.entropy() and
sdependent() with uvarpro().partialpro() gains a new vt.filter
argument for selecting the virtual-twin filtering engine. The default,
vt.filter = "isopro", preserves the existing
isolation-forest filtering behavior. New alternatives are
vt.filter = "outpro", which uses outpro-based
out-of-distribution support, and vt.filter = "none", which
disables VT filtering.outpro-based VT filtering to
partialpro(). For vt.filter = "outpro",
virtual twins are scored by an outpro distance, calibrated
against an outpro.null() reference distribution, and
converted to a support score. The existing cut option is
retained: larger values require stronger support and
cut = 0 disables VT filtering.distancef = "knn" to outpro(). The
KNN distance is computed in the standardized selected predictor subspace
and provides a faster option for large prediction or virtual-twin grids
because it does not require the forest-neighborhood distance
construction.outpro VT filter in partialpro() uses
KNN distance by default through the hidden option
out.distancef = "knn". Additional advanced controls are
available through ..., including out.neighbor,
out.reduce, out.cutoff,
out.max.rules.tree, out.max.tree,
out.knn.chunk.size, and out.null.outpro() now supports newdata.xscale,
allowing package-internal callers to pass new data that are already
aligned to the fitted VarPro x-scale. This is useful for functions such
as partialpro(), where virtual data are constructed
directly from the stored VarPro design matrix.outpro.null() now supports
nulldata.xscale, providing the corresponding x-scale option
for null/reference data.partialpro() help file with a fuller
description of the case-local partial-profile method, virtual-twin
filtering, local polynomial smoothing, classification log-odds handling,
binary-variable handling, and advanced options passed through
....outpro() documentation to describe the KNN
distance option and the x-scale handling used by package-internal
calls.partialpro() so that
nodesize is read from nodesize, not from
ntree.outpro.null() now uses cutoff = NULL by
default, matching the main outpro() cutoff-selection rule
and keeping null calibration consistent with ordinary
outpro() calls.importance() is now a true S3 generic rather than an
alias-style front end.partial.ivarpro() has been replaced by
plot.ivarpro().importance(),
predict(), and plot().importance() methods for
"varpro" and "uvarpro" objects.plot() methods for "ivarpro"
and "partialpro" objects.predict() methods
through standard S3 dispatch for "varpro",
"uvarpro", "ivarpro", and
"isopro" objects.plot.ivarpro, plot.partialpro,
predict.ivarpro, predict.varpro,
predict.uvarpro, and predict.isopro so that
method pages remain easy to find in the reference manual and via
?topic.\method{plot}{ivarpro}(x, ...)
and \method{predict}{ivarpro}(object, ...).plot(x, ...), predict(object, ...), and
importance(object).data for
the original feature matrix and documents target explicitly
for multivariate and multiclass outputs.partial.ivarpro(iv, var = ...) with
plot(iv, var = ...).importance(fit) over direct calls to
importance.varpro(fit).predict(fit, ...) over direct calls to
predict.class(fit, ...).varpro.strength() to reduce R-side
post-processing overhead after the native varProStrength
call, improving performance on large forests and large membership
reconstructions.varpro.strength(..., membership = TRUE) for very large
analyses.varpro.strength() now uses the
integrated hazard exposure values stored on the fitted object
(int.haz.oob) as the default working response when
available.cumsum() followed by a downstream missing-value error in
membership reconstruction.varpro() function.ivarPro.mclapply() with PSOCK-based
parallel execution, improving Windows compatibility.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.