prepCbnn: prepCbnn

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

prepCbnn

Description

This function attaches the required components to make survival outcomes possible from a binary classification network. Namely, it adds an offset term and passes the result through a Sigmoid activation function. This offset term is an offset in the statistical sense. As such, the network will add the offset term before passing through a sigmoid.

Usage

prepCbnn(
  features,
  nnInput,
  nnOutput,
  data,
  offset = NA,
  timeVar,
  eventVar,
  ratio = 100,
  compRisk = FALSE,
  kOptim = keras::optimizer_adam()
)

Arguments

features	(list) Input features to be used in model.
nnInput	(keras layer) Input layer for data to enter network.
nnOutput	(keras layer) Final keras layer of user designed network. Expected final layer is a single node.
data	(data.frame) Dataset to be used. can either be pre-casebase sampled with the sampleCaseBase function or the original dataset.
offset	(vector) Column of offset values generated from sampleCaseBase.
timeVar	(string) Survival time feature over which to eventually predict.
eventVar	(string) The event Feature in the data-set.
ratio	(numeric) Number of base-series samples per case-series samples. Default=100.
compRisk	(boolean) states if modeling competing risks or single event.
kOptim	("keras.optimizers.adam.Adam") Optimizer of choice from keras. Default set to optimizer_adam()

Value

(list) network: keras neural network model,
casebaseData: case-base sampled data,
offset: offset defined by case-base sampling,
timeVar: time variable (user defined),
eventVar: event variable (user defined),
features: feature of interest with user defined order.

Examples

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

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