In tidymodels/parsnip: A Common API to Modeling and Analysis Functions

r descr_models("rand_forest", "aorsf")

Tuning Parameters

defaults <-
  tibble::tibble(parsnip = c("trees", "min_n", "mtry"),
                 default = c("500L", "5L", "ceiling(sqrt(n_predictors))"))

param <-
  rand_forest() %>%
  set_engine("aorsf") %>%
  set_mode("censored regression") %>%
  make_parameter_list(defaults) %>%
  distinct()

This model has r nrow(param) tuning parameters:

param$item

Additionally, this model has one engine-specific tuning parameter:

• split_min_stat: Minimum test statistic required to split a node. Defaults are 3.841459 for censored regression (which is roughly a p-value of 0.05) and 0 for classification and regression. For classification, this tuning parameter should be between 0 and 1, and for regression it should be greater than or equal to 0. Higher values of this parameter cause trees grown by aorsf to have less depth.

Translation from parsnip to the original package (censored regression)

r uses_extension("rand_forest", "aorsf", "censored regression")

library(censored)

rand_forest() %>%
  set_engine("aorsf") %>%
  set_mode("censored regression") %>%
  translate()

Translation from parsnip to the original package (regression)

r uses_extension("rand_forest", "aorsf", "regression")

library(bonsai)

rand_forest() %>%
  set_engine("aorsf") %>%
  set_mode("regression") %>%
  translate()

Translation from parsnip to the original package (classification)

r uses_extension("rand_forest", "aorsf", "classification")

library(bonsai)

rand_forest() %>%
  set_engine("aorsf") %>%
  set_mode("classification") %>%
  translate()

Preprocessing requirements

Case weights

Other details

Predictions of survival probability at a time exceeding the maximum observed event time are the predicted survival probability at the maximum observed time in the training data.

The class predict method in aorsf uses the standard 'each tree gets one vote' approach, which is usually but not always consistent with the picking the class that has highest predicted probability. It is okay for this inconsistency to occur in aorsf because it is intentionally applying the traditional class prediction method for random forests, but in tidymodels it is preferable to embrace consistency. Thus, we opted to make predicted probability consistent with predicted class all the time by making the predicted class a function of predicted probability (see tidymodels/bonsai#78).

References

• Jaeger BC, Long DL, Long DM, Sims M, Szychowski JM, Min YI, Mcclure LA, Howard G, Simon N. Oblique random survival forests. Annals of applied statistics 2019 Sep; 13(3):1847-83. DOI: 10.1214/19-AOAS1261

• Jaeger BC, Welden S, Lenoir K, Pajewski NM. aorsf: An R package for supervised learning using the oblique random survival forest. Journal of Open Source Software 2022, 7(77), 1 4705. \doi{10.21105/joss.04705}.

• Jaeger BC, Welden S, Lenoir K, Speiser JL, Segar MW, Pandey A, Pajewski NM. Accelerated and interpretable oblique random survival forests. arXiv e-prints 2022 Aug; arXiv-2208. URL: https://arxiv.org/abs/2208.01129

tidymodels/parsnip documentation built on Feb. 19, 2025, 2:10 a.m.