Survival PC-Hazard Learner

Logistic-Hazard fits a discrete neural network based on a cross-entropy loss and predictions of a discrete hazard function, also known as Nnet-Survival. Calls survivalmodels::pchazard() from package 'survivalmodels'.

Details

Custom nets can be used in this learner either using the survivalmodels::build_pytorch_net utility function or using torch via reticulate. The number of output channels depends on the number of discretised time-points, i.e. the parameters cuts or cutpoints.

Dictionary

This Learner can be instantiated via the dictionary mlr_learners or with the associated sugar function lrn():

mlr_learners$get("surv.pchazard")
lrn("surv.pchazard")

Meta Information

• Task type: “surv”

• Predict Types: “crank”, “distr”

• Feature Types: “integer”, “numeric”

• Required Packages: mlr3, mlr3proba, mlr3extralearners, survivalmodels, distr6, reticulate

Parameters

Id	Type	Default	Levels	Range
frac	numeric	0		\([0, 1]\)
cuts	integer	10		\([1, \infty)\)
cutpoints	untyped	-		-
scheme	character	equidistant	equidistant, quantiles	-
cut_min	numeric	0		\([0, \infty)\)
num_nodes	untyped	c , 32, 32		-
batch_norm	logical	TRUE	TRUE, FALSE	-
reduction	character	mean	mean, none, sum	-
dropout	numeric	-		\([0, 1]\)
activation	character	relu	celu, elu, gelu, glu, hardshrink, hardsigmoid, hardswish, hardtanh, relu6, leakyrelu, ...	-
custom_net	untyped	-		-
device	untyped	-		-
optimizer	character	adam	adadelta, adagrad, adam, adamax, adamw, asgd, rmsprop, rprop, sgd, sparse_adam	-
rho	numeric	0.9		\((-\infty, \infty)\)
eps	numeric	1e-08		\((-\infty, \infty)\)
lr	numeric	1		\((-\infty, \infty)\)
weight_decay	numeric	0		\((-\infty, \infty)\)
learning_rate	numeric	0.01		\((-\infty, \infty)\)
lr_decay	numeric	0		\((-\infty, \infty)\)
betas	untyped	c , 0.9 , 0.999		-
amsgrad	logical	FALSE	TRUE, FALSE	-
lambd	numeric	1e-04		\([0, \infty)\)
alpha	numeric	0.75		\([0, \infty)\)
t0	numeric	1e+06		\((-\infty, \infty)\)
momentum	numeric	0		\((-\infty, \infty)\)
centered	logical	TRUE	TRUE, FALSE	-
etas	untyped	c , 0.5, 1.2		-
step_sizes	untyped	c , 1e-06, 50		-
dampening	numeric	0		\((-\infty, \infty)\)
nesterov	logical	FALSE	TRUE, FALSE	-
batch_size	integer	256		\((-\infty, \infty)\)
epochs	integer	1		\([1, \infty)\)
verbose	logical	TRUE	TRUE, FALSE	-
num_workers	integer	0		\((-\infty, \infty)\)
shuffle	logical	TRUE	TRUE, FALSE	-
best_weights	logical	FALSE	TRUE, FALSE	-
early_stopping	logical	FALSE	TRUE, FALSE	-
min_delta	numeric	0		\((-\infty, \infty)\)
patience	integer	10		\((-\infty, \infty)\)
interpolate	logical	FALSE	TRUE, FALSE	-
sub	integer	10		\([1, \infty)\)

Installation

Package 'survivalmodels' is not on CRAN and has to be install from GitHub via remotes::install_github("RaphaelS1/survivalmodels").

References

Kvamme, Håvard, Borgan Ø (2019). “Continuous and discrete-time survival prediction with neural networks.” arXiv preprint arXiv:1910.06724.

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

RaphaelS1

Super classes

mlr3::Learner -> mlr3proba::LearnerSurv -> LearnerSurvPCHazard

Methods

Public methods

• LearnerSurvPCHazard$new()

• LearnerSurvPCHazard$clone()

Inherited methods

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

---

Method new()

Creates a new instance of this R6 class.

Usage

LearnerSurvPCHazard$new()

---

Method clone()

The objects of this class are cloneable with this method.

Usage

LearnerSurvPCHazard$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

Examples

learner = mlr3::lrn("surv.pchazard")
print(learner)
#> <LearnerSurvPCHazard:surv.pchazard>: PC-Hazard Learner
#> * Model: -
#> * Parameters: list()
#> * Packages: mlr3, mlr3proba, mlr3extralearners, survivalmodels, distr6,
#>   reticulate
#> * Predict Type: crank
#> * Feature types: integer, numeric
#> * Properties: -

# available parameters:
learner$param_set$ids()
#>  [1] "frac"           "cuts"           "cutpoints"      "scheme"        
#>  [5] "cut_min"        "num_nodes"      "batch_norm"     "reduction"     
#>  [9] "dropout"        "activation"     "custom_net"     "device"        
#> [13] "optimizer"      "rho"            "eps"            "lr"            
#> [17] "weight_decay"   "learning_rate"  "lr_decay"       "betas"         
#> [21] "amsgrad"        "lambd"          "alpha"          "t0"            
#> [25] "momentum"       "centered"       "etas"           "step_sizes"    
#> [29] "dampening"      "nesterov"       "batch_size"     "epochs"        
#> [33] "verbose"        "num_workers"    "shuffle"        "best_weights"  
#> [37] "early_stopping" "min_delta"      "patience"       "interpolate"   
#> [41] "sub"