Changes in Version 0.5 (2018-05-07)
- Function pre() now also supports multinomial and multivariate
gaussian response variables.
- Function pre() now has argument 'tree.unbiased'; if set to FALSE,
the CART algorithm (as implemented in package 'rpart') is employed
for rule induction.
- Argument 'maxdepth' of function pre() allows for specifying
varying maximum depth across trees, through specifying a
vector of length ntrees, or a random number generating function.
See ?maxdepth.sampler for details.
Changes in Version 0.4 (2017-08-31)
- Added dataset 'carrillo'
- By default, a gradient boosting approach is now taken for all response
types. That is, partykit::ctree() and a learning rate of .01 is
employed by default. Alternatively, glmtree() can be employed for tree
induction by sprecifying use.grad = FALSE.
- The 'family' argument in pre() now takes character strings as well as glm
family objects.
- Functions pairplot() and interact() now use HCL instead of highly saturated
HSV colors as default plotting colors.
- Bug fixed in plot.pre: Node directions are now in accordance with
rule definition.
- Bug fixed in predict.pre: No error printed when response variable is not
supplied.
Changes in Version 0.3 (2017-08-03):
- Function gpe() added, which fits general prediction ensembles. By default,
it fits an ensemble of rules, linear and hinge functions. Function gpe()
allows for specifying custom baselearner generating functions and a custom
fitting function for the final model.
- Numerous bugs fixed, yielding faster computation times and clearer plots
with more customization options.
- Added support for count responses. Function pre() now has a 'family'
argument, which should be set to 'poisson' for count outcomes (the
'family' argument is set automatically to 'gaussian' for numeric response
variables and to 'binomial' for binary response variables (factors)).
- A gradient boosting approach for binary outcomes is applied, by default,
substantially reducing computation times. This can be turned off through
the 'use.grad' argument in function pre().
- The default of the 'learnrate' argument of function pre() has been changed
to .01, by default. Before, it was .01 for continuous outcomes, but 0 for
binary outcomes, to reduce computation time. With gradient boosting
implemented, computation time is much reduced.
- Argument 'tree.control' in function pre() allows for passing arguments to
partykit tree fitting functions.
- Arguments for the cv.glmnet() function are directly passed through better
use of ... . Most importantly, this means that argument 'mod.sel.crit'
cannot be used anymore and should be referred to as 'type.measure'
(which will be directly passed to cv.glmnet). Similarly, 'thres' and
'standardize' are not explicit arguments of function pre() anymore and
can now be directly passed to cv.glmnet() using ... .
- Better use of sample weights: weights specified with the 'weights' argument
in pre() are now used as weights in the subsampling procedure, instead of
as observation weights in the tree-fitting procedure.
- Added corplot() function, which shows the correlation between the
baselearners in the ensemble.
- Function pairplot() returns a heatmap by default, a 3D or contour plot can
also be requested.
- Appearance of plot resulting from interaction() improved.
Changes in Version 0.2 (2017-04-25):
- Added print() and plot() method for objects of class pre.
- Added support for using functions like factor() and log() in formula
statement of function pre(). (thanks to Bill Venables for suggesting this)
- Added support for parallel computating in functions pre(), cvpre(),
bsnullinteract() and interact().
- Winsorizing points used for the linear terms are reported in the description
of the base learners returned by coef() and importance(). (Thanks to
Rishi Sadhir for suggesting this)
- Added README file.
- Legend included in plot for interaction test statistics.
- Fixed importance() function to allow for selecting final ensemble with
different value than 'lambda.1se'.
- Cleaned up all occurrences of set.seed()
- Fixed cvpre() function: penalty.par.val argument now included
- Many minor bug fixes.
Changes in Version 0.1 (2016-12-23):
- First CRAN release.