Updated the list.rules function in pre and gpe to give all but one of the two root rules from the trees
Removed wrappers for ctree
Simplified tests so the run faster. You will need to source tests/helper.R when you make tests in the future. See ?testthat::source_file
All tests passes with partykit-1.2.0 (Run install.packages("partykit", repos="http://R-Forge.R-project.org")) . Some failing test can be avoided temporarily that fails with partykit-1.1.1. Change tree.control = ctree_control(mtry = mtry, maxdepth = maxdepth, saveinfo= FALSE) to tree.control = ctree_control(mtry = mtry, maxdepth = maxdepth) in gpe and then all tests passes
gpe has almost the same defaults as pre
predict.pre also winsorize. One function is now used to make model.matrix in pre which I find simpler to grasp
The latter means that this test passes:
test_that("Predict gives previous results with airquality data", {
set.seed(42)
airq.ens <- pre(Ozone ~ ., data=airquality, ntrees = 10)
preds <- predict(airq.ens, airquality)
#####
# Should give the same if the same data is passed to newdata as used to fit
other_preds <- predict(airq.ens)
expect_equal(other_preds, preds)
})
In other words, you get the same predicted results for the initial data both with the function and by calling predict. Further, the new function to get the model.matrix will also work if the user, for some reason, have dropped the factor levels for one newdata's columns in the argument to predict and other edge cases.
Changes
gpeand factorslist.rulesfunction inpreandgpeto give all but one of the two root rules from the treesctreetests/helper.Rwhen you make tests in the future. See?testthat::source_fileinstall.packages("partykit", repos="http://R-Forge.R-project.org")) . Some failing test can be avoided temporarily that fails with partykit-1.1.1. Changetree.control = ctree_control(mtry = mtry, maxdepth = maxdepth, saveinfo= FALSE)totree.control = ctree_control(mtry = mtry, maxdepth = maxdepth)ingpeand then all tests passesgpehas almost the same defaults asprepredict.prealso winsorize. One function is now used to makemodel.matrixinprewhich I find simpler to graspThe latter means that this test passes:
In other words, you get the same predicted results for the initial data both with the function and by calling
predict. Further, the new function to get themodel.matrixwill also work if the user, for some reason, have dropped the factor levels for onenewdata's columns in the argument topredictand other edge cases.