SurvDNN Survival Learner

Deep neural network survival models from package survdnn, aimed at tabular (low to moderate-dimensional) covariate settings using torch-based multilayer perceptrons. The learner wraps survdnn::survdnn().

Prediction types

This learner supports the following prediction types:

lp

A numeric vector of linear predictors, one per observation. For loss "cox" / "cox_l2" this is a log-risk score (higher implies worse prognosis). For "aft", predict.survdnn() returns the predicted log-time location \(\mu(x)\) (higher implies better prognosis), therefore the learner internally negates it such that higher values imply higher risk (consistent with mlr3 conventions). For "coxtime", this is \(g(t_0, x)\) evaluated at a reference time.

crank

same as lp.

distr

A survival matrix (rows = observations, columns = time points) based on predict(type = "survival"). By default, predictions are evaluated on the unique event times of the training data.

Dictionary

This Learner can be instantiated via lrn():

lrn("surv.survdnn")

Meta Information

• Task type: “surv”

• Predict Types: “crank”, “distr”, “lp”

• Feature Types: “integer”, “numeric”, “factor”, “ordered”

• Required Packages: mlr3, mlr3proba, mlr3extralearners, survdnn, torch

Parameters

Id	Type	Default	Levels	Range
hidden	untyped	c(32L, 16L)		-
activation	character	relu	relu, leaky_relu, tanh, sigmoid, gelu, elu, softplus	-
lr	numeric	1e-04		\([1e-06, 1]\)
epochs	integer	300		\([1, \infty)\)
loss	character	cox	cox, cox_l2, aft, coxtime	-
optimizer	character	adam	adam, adamw, sgd, rmsprop, adagrad	-
optim_args	untyped	list()		-
verbose	logical	FALSE	TRUE, FALSE	-
dropout	numeric	0.3		\([0, 1]\)
batch_norm	logical	TRUE	TRUE, FALSE	-
callbacks	untyped	NULL		-
.seed	integer	NULL		\((-\infty, \infty)\)
.device	character	auto	auto, cpu, cuda	-
na_action	character	omit	omit, fail	-

See also

• Dictionary of Learners: mlr3::mlr_learners.

• as.data.table(mlr_learners) for a table of available Learners in the running session (depending on the loaded packages).

• Chapter in the mlr3book: https://mlr3book.mlr-org.com/basics.html#learners

• mlr3learners for a selection of recommended learners.

• mlr3cluster for unsupervised clustering learners.

• mlr3pipelines to combine learners with pre- and postprocessing steps.

• mlr3tuning for tuning of hyperparameters, mlr3tuningspaces for established default tuning spaces.

Author

ielbadisy

Super classes

mlr3::Learner -> mlr3proba::LearnerSurv -> LearnerSurvDNN

Active bindings

marshaled

(logical(1)) Whether the learner has been marshaled.

Methods

Public methods

• LearnerSurvDNN$new()

• LearnerSurvDNN$marshal()

• LearnerSurvDNN$unmarshal()

• LearnerSurvDNN$clone()

Inherited methods

• mlr3::Learner$base_learner()
• mlr3::Learner$configure()
• mlr3::Learner$encapsulate()
• mlr3::Learner$format()
• mlr3::Learner$help()
• mlr3::Learner$predict()
• mlr3::Learner$predict_newdata()
• mlr3::Learner$print()
• mlr3::Learner$reset()
• mlr3::Learner$selected_features()
• mlr3::Learner$train()

---

Method new()

Creates a new instance of this R6 class.

Usage

LearnerSurvDNN$new()

---

Method marshal()

Marshal the learner's model.

Usage

LearnerSurvDNN$marshal(...)

Arguments

...

(any)
Additional arguments passed to mlr3::marshal_model().

---

Method unmarshal()

Unmarshal the learner's model.

Usage

LearnerSurvDNN$unmarshal(...)

Arguments

...

(any)
Additional arguments passed to mlr3::unmarshal_model().

---

Method clone()

The objects of this class are cloneable with this method.

Usage

LearnerSurvDNN$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

Examples

# Define the Learner
learner = lrn("surv.survdnn", epochs = 42L, loss = "cox")
print(learner)
#> 
#> ── <LearnerSurvDNN> (surv.survdnn): SurvDNN (torch-based deep survival models) ─
#> • Model: -
#> • Parameters: epochs=42, loss=cox
#> • Packages: mlr3, mlr3proba, mlr3extralearners, survdnn, and torch
#> • Predict Types: [crank], distr, and lp
#> • Feature Types: integer, numeric, factor, and ordered
#> • Encapsulation: none (fallback: -)
#> • Properties: marshal
#> • Other settings: use_weights = 'error'

# Define the task, split to train/test set
task = tsk("lung")
set.seed(42)
part = partition(task)

# Train the learner on the training ids
learner$train(task, row_ids = part$train)
#> Error: The Torch backend is not installed.
#> Please run: torch::install_torch().
print(learner$model)
#> NULL

# Make predictions for the test rows
predictions = learner$predict(task, row_ids = part$test)
#> Error: 
#> ✖ Cannot predict, Learner 'surv.survdnn' has not been trained yet
#> → Class: Mlr3ErrorInput
print(predictions)
#> Error: object 'predictions' not found

# Score the predictions
predictions$score()
#> Error: object 'predictions' not found