Accelerated oblique random forests

The aorsf package is a software project dedicated to oblique random forests. It was originally the obliqueRSF package introduced in the oblique random survival forest paper, and had trouble with inefficiency and lack of features for interpretation. Today, it has grown into an R package that efficiently fits oblique random forests for classification, regression, and survival, with customized features to interpret them.

When aorsf was reviewed by R OpenSci, one of my reviewers was Marvin N. Wright, author of the ranger package. At the time, aorsf was only designed to fit oblique random survival forests, and one of Marvin’s review comments was

What about oblique RF for classification and regression? Do other packages implement the speedups used here? If not, why not have a fast oblique RF package for all outcomes instead of a specific one for survival?

It was a great question. Over the years there have been packages that provide specific types of oblique random forests, like obliqueRF, RLT, rotationForest, and others. But, as far as I know, there was never an R package that brought oblique random forests together.

library(aorsf)
library(obliqueRSF)
library(survival)
library(microbenchmark)

options(microbenchmark.unit="relative")

flc <- na.omit(survival::flchain)
flc$futime <- pmax(flc$futime, 0.01)

microbenchmark(
 obliqueRSF = ORSF(data = flc, 
                   time = 'futime', 
                   status = 'death', 
                   ntree = 10, 
                   verbose = FALSE),
 aorsf = orsf(futime + death ~ ., 
              data = flc, 
              n_tree = 10),
 times = 1
)

Unit: relative
       expr      min       lq     mean   median       uq      max neval
 obliqueRSF 245.9676 245.9676 245.9676 245.9676 245.9676 245.9676     1
      aorsf   1.0000   1.0000   1.0000   1.0000   1.0000   1.0000     1