| NEWS | R Documentation |
min_update is the number of observations with non-zero weight.
Documentation was correct but code worked with sum(weights).
Documention and test updates.
New intercept argument for residual-based tree induction (shift and scale).
Plot discrete distributions.
Remove maxit argument (was ignored by mlt).
Replace .R2vec by as.double method.
Update references.
Setting min\_update might have caused constant predictions.
Update simulations / demos.
Add min\_update parameter; model parameters are only
updated when the number of observations in a node exceeds this number.
Defaults to two times the number of parameters in the model.
Failure to update model parameters lead to a stop in tree
recursion. Following a remark by Stefan Wager, the parameters
from the parent node are now used in such a case, allowing
larger trees to be grown by keeping the scores constant.
Note that update = FALSE triggers this behaviour
in all nodes below the root node.
Add R code reproducing empirical results for "Survival Forests under Test".
Respect scale argument from object.
traforest broke in extremely rare cases due to incorrect
dimensions of the score matrix.
All reparameterisation for max-type tests; this might give better (more powerful) variable selection.
Allow fixed coefficients in trees and forests; this is useful for log-rank splitting, for example.
Document predict and logLik methods.
Add applyfun and cores arguments to predict
method for transformation forests (allowing parallel computation
of predictions.
Some improvements in internal functions.
Simulations with 500 (not 50) noise variables.
Simplify likelihood evaluations based on new capabilities in mlt.
New update argument for transformation forests.
Update outputs (scale = TRUE in partykit 1.2-1).
The trtf package was first published on CRAN.