NA

This document list all the changes that are made to the source code of rpart to tranfsorm the package to distRforest.

Name change from rpart to distRforest

• DESCRIPTION
  • update package name, title, description, url,…
• NAMESPACE
  • line 1: change rpart to distRforest in useDynLib(distRforest, .registration = TRUE, .fixes = "C_")
• src/init.c
  • line 23: change rpart to distRforest in R_init_distRforest(DllInfo * dll)
• R/rpart.R
  • line 78: change rpart to distRforest in ns <- asNamespace("distRforest")
• R/zzz.R
  • line 1: change rpart to distRforest in library.dynam.unload("distRforest", libpath)

Random forest functionality

• src/rpartproto.h
  • line 39: add SEXP ncand2 and SEXP seed2 as input arguments to SEXP rpart(...)
  • line 64: add void sample(int *vector, int x, int size);
• src/rpart.h
  • line 43: add int ncand; and int seed; to EXTERN struct {...} rp;
• src/rpart.c
  • line 41: add SEXP ncand2 and SEXP seed2 as input arguments to rpart(...)
  • line 116-117: add both to the rp struct by rp.ncand = asInteger(ncand2); and rp.seed = asInteger(seed2);
  • line 119: set the seed via srand(rp.seed);
• R/rpart.R
  • line 4: add ncand, seed and redmem = FALSE as input arguments to rpart <- function(...)
  • line 9: add if (missing(parms)) parms <- NULL
  • line 50 and 71: change condition missing(parms) to is.null(parms)
  • line 153-154: set ncand equal to number of available variables when input argument is missing
  • line 157: set seed equalt to 1 when input argument is missing
  • line 163: add as.integer(ncand) and as.integer(seed) as input arguments to rpfit <- .Call(C_rpart,...)
  • line 301: add if (redmem) for(x in c(...)) ans[[x]] <- NULL to reduce memory of fitted rpart object
• src/init.c
  • line 14: change the number of input parameters from 11 to 13 in {"rpart", (DL_FUNC) &rpart, 13}
• src/sample.c
  • create this file to be able to sample in the split candidates
• src/bsplit.c
  • line 39: add int candidates[rp.ncand]; to allocate a vector to store the split candidates in
  • line 40: add sample(candidates, rp.nvar, rp.ncand); to sample the split candidates
  • line 50: change rp.nvar to rp.ncand in for (i = 0; i < rp.ncand; i++) to iterate over subset
  • line 51-94: change i to candidates[i] in for body to select correct split candidate from the sample

User-friendly random forest functions

• R/rforest.R
  • add this file to make it easier for the user to build a random forest
• R/predict.rforest.R
  • add this file to make it easier for the user to predict from a random forest
• R/importance_rforest.R
  • add this file to make it easier for the user to calculate variable importance scores from a random forest
• NAMESPACE
  • line 7: add rforest, predict.rforest and importance_rforest to export(...)
• r/error_funcs.R
  • add this file containing the functions to calculate the OOB error for the different method options

Log-normal and gamma distributions

• src/anova.c
  • line 191-196: merge anovapred.c file in anova.c as the anovapred(...) function
• src/lognormal.c
  • add this file containing the functions necessary to support lognormal splitting
• src/gamma.c
  • add this file containing the functions necessary to support gamma splitting
• src/func_table.h
  • line 21-24: add extern int lognormalinit(...) and extern int gammainit(...)
  • line 35-38: add extern int lognormaldev(...) and extern int gammadev(...)
  • line 52-57: add extern int lognormal(...) and extern int gammasplit(...)
  • line 63-64: add extern int lognormalpred(...) and extern int gammapred(...)
• R/rpart.lognormal.R
  • add this file to initialize the lognormal tree and check whether the response input is strictly positive
• R/rpart.gamma.R
  • add this file to initialize the gamma tree and check whether the response input is strictly positive
• R/rpart.R
  • line 59 and 61: add "lognormal" and "gamma" to vector of possible methods
  • line 109: add warning that CV is not supported when method %in% c('lognormal','gamma') && xval > 0L

Miscellaneous

• vignettes/
  • remove original rpart vignettes in the distRforest package
  • add vignette on the use of distRforest with examples for classification, Poisson and gamma regression
• tests/
  • remove original rpart tests in the distRforest package
  • test-distributions.R: test the method = 'gamma' and method = 'lognormal' options added to rpart(...)
  • test-rforest.R: test whether rforest() is able to replicate identical trees as obtained with rpart()
  • test-outofbag.R: test if the tracking of the out-of-bag error with rforest() happens as expected
  • test-predict.R: test if predictions with predict.rforest() are being generated as expected
  • test-importance.R: test if importance_rforest() calculates the variable importance scores as expected