Fix "donttest" examples, broken by partykit 1.3-0.
Import functions selectively, as needed, from graphics and partykit
rather than the entire namespaces.
Code repository changed from R-Forge to Codeberg at: https://codeberg.org/zeileis/stablelearner/
Added basic altdoc page with overview and documentation at: https://zeileis.codeberg.page/stablelearner/
splithalf() sampling for stability(). In case of
an uneven number of observations in the learning sample some small overlap
in the two splitted halves would occur. This is avoided now by dropping
one random observation now and restricting both halves to be always of
the same size. Reported by Constantin Wiegand along with two further
small improvements: Warnings about lack of variance in predictions on
learning samples and avoiding manual triggering of the garbage collector.The as.stabletree() method for RandomForest objects (party package)
is now registered as an S3 (rather than S4) method.
Environment .stabEnv is only used internally by stabletree() and
hence not exported anymore in NAMESPACE.
Z takes over maintenance from MP.
Fix checks for sampler specifications where a logical of length 2 is
now correctly aggregated with all().
Changed default sampling method in stabletree() from bootstrap() to
subsampling() with default fraction of v = 0.632.
as.stabletree() coercion generic added which allows to coerce a
randomForest (randomForest package), RandomForest (party package),
cforest (partykit package) or ranger (ranger package) to a
stabletree object.
Added a vignette on the variable and cutpoint selection analysis of random forests.
Project stablelearner has been launched and a stable version of the
package has been uploaded to CRAN.
stability() is available to estimate the stability of the results
from a given supervised statistical learning method.
stabletree() is available to estimate the stability of the results
from recursive partitioning.