What is survivalmodels?
survivalmodels implements models for survival analysis that are either not already implemented in R, or novel implementations for speed improvements. Currently implemented are five neural networks from the Python packages pycox, DNNSurv, and the Akritas non-parametric conditional estimator. Further updates will include implementations of novel survival models.
For a hands-on demonstration of model training, tuning, and comparison see this article I wrote, which uses the mlr3proba interface with models from survivalmodels.
Basic Usage
# load dependencies
library(survival)
train <- simsurvdata(100)
test <- simsurvdata(50)
fit <- akritas(Surv(time, status) ~ ., data = train)
predict(fit, newdata = test)
# Use distr6 = TRUE to return a distribution
predict_distr <- predict(fit, newdata = test, distr6 = TRUE)
predict_distr$survival(100)
# Return a relative risk ranking with type = "risk"
predict(fit, newdata = test, type = "risk")
Or both survival probabilities and a rank
predict(fit, newdata = test, type = "all", distr6 = TRUE)Python Models
survivalmodels implements models from Python using reticulate. In order to use these models, the required Python packages must be installed following with reticulate::py_install. survivalmodels includes a helper function to install the required pycox function (with pytorch if also required). Before running any models in this package, if you have not already installed pycox please run
install_pycox(pip = TRUE, install_torch = FALSE)With the arguments changed as you require, see ?install_pycox for more.
For DNNSurv the model depends on keras and tensorflow, which require installation via:
install_keras(pip = TRUE, install_tensorflow = FALSE)Installation
Install the latest release from CRAN:
install.packages("survivalmodels")Install the development version from GitHub:
remotes::install_github("RaphaelS1/survivalmodels")