distr prediction is estimated using the fitted custom distributions
flexsurv::flexsurvspline() and the estimated coefficients however the prediction takes
place in this package and not in flexsurv for a much faster and more efficient
As flexible spline models estimate the baseline hazard as the intercept, the linear predictor,
lp, can be calculated as in the classical setting. i.e. For fitted coefficients,
\(\beta = (\beta_0,...,\beta_P)\),
and covariates \(X^T = (X_0,...,X_P)^T\), where \(X_0\) is a column
of \(1\)s: \(lp = \beta X\).
Packages: flexsurv, pracma
Predict Types: distr, crank, lp
Feature Types: logical, integer, factor, numeric
Reason for change: The default value of
0 is equivalent to, and a much less efficient
implementation of, LearnerSurvParametric.
Royston P, Parmar MKB (2002). “Flexible parametric proportional-hazards and proportional-odds models for censored survival data, with application to prognostic modelling and estimation of treatment effects.” Statistics in Medicine, 21(15), 2175–2197. doi: 10.1002/sim.1203.
Creates a new instance of this R6 class.
The objects of this class are cloneable with this method.
LearnerSurvFlexible$clone(deep = FALSE)
Whether to make a deep clone.
# stop example failing with warning if package not installed learner = suppressWarnings(mlr3::lrn("surv.flexible")) print(learner)#> <LearnerSurvFlexible:surv.flexible> #> * Model: - #> * Parameters: k=1 #> * Packages: flexsurv, pracma #> * Predict Type: distr #> * Feature types: logical, integer, factor, numeric #> * Properties: weights# available parameters: learner$param_set$ids()#>  "bhazard" "k" "knots" "bknots" #>  "scale" "timescale" "inits" "rtrunc" #>  "fixedpars" "cl" "maxiter" "rel.tolerance" #>  "toler.chol" "debug" "outer.max"