README.md

In Jesse-Islam/cbnn: Case-Base Neural networks

cbnn

R package for case-base neural networks.

This package is based on keras. As its R implementation has python-based dependencies, its important to follow https://tensorflow.rstudio.com/install/ and correctly install keras. Then, install the cbnn package

#follow https://tensorflow.rstudio.com/install/
#install.packages('devtools')
devtools::install_github("Jesse-Islam/cbnn")

You can then run the following code to make sure everything installed correctly. If there are any problems, feel free to contact me on this repository and open an issue.

#load packages
library(cbnn)
library(casebase)
library(magrittr)

#process data
data<-casebase::ERSPC
data$ScrArm<-as.numeric(data$ScrArm)-1
eventVar<-"DeadOfPrCa"
timeVar<-"Follow.Up.Time"
features<-"ScrArm"

#prepare CBNN model for fitting
nnInput<-keras::layer_input(shape=length(features))
nnOutput<-nnInput %>% keras::layer_dense(units=1, use_bias = TRUE)
cbnnPrep<-prepCbnn(features, nnInput, nnOutput, data, offset=NA,timeVar,
eventVar, ratio=10, compRisk=FALSE)

#Fit model
fit<-fitHazard(cbnnPrep,epochs=1,batchSize=500)

#calculate cumulative incidence curves
times<-seq(min(data$Follow.Up.Time),max(data$Follow.Up.Time),length.out=10)
x_test<-as.matrix(data[sample(nrow(data),100),])
cumulativeIncidenceCurves<-cuIncCbnn(fit, times=times,x_test=x_test)

I recommend going through the vignette (https://github.com/Jesse-Islam/cbnn/tree/main/doc/Time-varying-interactions-and-flexible-baseline-hazard.html) next, as it uses the complex simulation from the manuscript ( https://github.com/Jesse-Islam/cbnnManuscript ) as an example to better understand the implementation of CBNN on tabular data.

If you're interested in reproducing the results from the paper, the exact scripts used to generate the figures are available in https://github.com/Jesse-Islam/cbnnManuscript/tree/main/analyses.

Please cite these papers if you use this method in your work:

   Islam, J., Turgeon, M., Sladek, R., & Bhatnagar, S. (2023).
   Case-Base Neural Networks: survival analysis with time-varying,
   higher-order interactions. arXiv preprint arXiv:2301.06535.

   Bhatnagar S, Turgeon M, Islam J, Saarela O, Hanley J (2020).
   _casebase: Fitting Flexible Smooth-in-Time Hazards and Risk Functions
   via Logistic and Multinomial Regression_. R package version 0.9.0,
   <https://CRAN.R-project.org/package=casebase>.

   Hanley, James A., and Olli S. Miettinen. Fitting smooth-in-time
   prognostic risk functions via logistic regression. International
   Journal of Biostatistics 5.1 (2009): 1125-1125.

   Saarela, Olli. A case-base sampling method for estimating recurrent
   event intensities. Lifetime data analysis 22.4 (2016): 589-605.

Jesse-Islam/cbnn documentation built on Jan. 13, 2024, 3:48 a.m.