survdnn: Fit a Deep Neural Network for Survival Analysis
In survdnn: Deep Neural Networks for Survival Analysis with R 'torch'

survdnn

R Documentation

Fit a Deep Neural Network for Survival Analysis

Description

Trains a deep neural network (DNN) to model right-censored survival data using one of the predefined loss functions: Cox, AFT, or Coxtime.

Usage

survdnn(
  formula,
  data,
  hidden = c(32L, 16L),
  activation = "relu",
  lr = 1e-04,
  epochs = 300L,
  loss = c("cox", "cox_l2", "aft", "coxtime"),
  optimizer = c("adam", "adamw", "sgd", "rmsprop", "adagrad"),
  optim_args = list(),
  verbose = TRUE,
  dropout = 0.3,
  batch_norm = TRUE,
  callbacks = NULL,
  .seed = NULL,
  .device = c("auto", "cpu", "cuda"),
  na_action = c("omit", "fail")
)

Arguments

`formula`	A survival formula of the form 'Surv(time, status) ~ predictors'.
`data`	A data frame containing the variables in the model.
`hidden`	Integer vector. Sizes of the hidden layers (default: c(32, 16)).
`activation`	Character string specifying the activation function to use in each layer. Supported options: '"relu"', '"leaky_relu"', '"tanh"', '"sigmoid"', '"gelu"', '"elu"', '"softplus"'.
`lr`	Learning rate for the optimizer (default: '1e-4').
`epochs`	Number of training epochs (default: 300).
`loss`	Character name of the loss function to use. One of '"cox"', '"cox_l2"', '"aft"', or '"coxtime"'.
`optimizer`	Character string specifying the optimizer to use. One of '"adam"', '"adamw"', '"sgd"', '"rmsprop"', or '"adagrad"'. Defaults to '"adam"'.
`optim_args`	Optional named list of additional arguments passed to the underlying torch optimizer (e.g., 'list(weight_decay = 1e-4, momentum = 0.9)').
`verbose`	Logical; whether to print loss progress every 50 epochs (default: TRUE).
`dropout`	Numeric between 0 and 1. Dropout rate applied after each hidden layer (default = 0.3). Set to 0 to disable dropout.
`batch_norm`	Logical; whether to add batch normalization after each hidden linear layer (default = TRUE).
`callbacks`	Optional list of callback functions. Each callback should have signature 'function(epoch, current)' and return TRUE if training should stop, FALSE otherwise. Used, for example, with [callback_early_stopping()].
`.seed`	Optional integer. If provided, sets both R and torch random seeds for reproducible weight initialization, shuffling, and dropout.
`.device`	Character string indicating the computation device. One of '"auto"', '"cpu"', or '"cuda"'. '"auto"' uses CUDA if available, otherwise falls back to CPU.
`na_action`	Character. How to handle missing values in the model variables: '"omit"' drops incomplete rows (and reports how many were removed when 'verbose=TRUE'); '"fail"' stops with an error if any missing values are present.

Value

An object of class '"survdnn"' containing:

model: Trained 'nn_module' object.
formula: Original survival formula.
data: Training data used for fitting.
xnames: Predictor variable names.
x_center: Column means of predictors.
x_scale: Column standard deviations of predictors.
loss_history: Vector of loss values per epoch.
final_loss: Final training loss.
loss: Loss function name used ("cox", "aft", etc.).
activation: Activation function used.
hidden: Hidden layer sizes.
lr: Learning rate.
epochs: Number of training epochs.
optimizer: Optimizer name used.
optim_args: List of optimizer arguments used.
device: Torch device used for training ('torch_device').
aft_log_sigma: Learned global log(sigma) for 'loss="aft"'; 'NA_real_' otherwise.
aft_loc: AFT log-time location offset used for centering when 'loss="aft"'; 'NA_real_' otherwise.
coxtime_time_center: Mean used to scale time for CoxTime; 'NA_real_' otherwise.
coxtime_time_scale: SD used to scale time for CoxTime; 'NA_real_' otherwise.