`multi_predict()` for `mboost` models

[hfrick]

A more specific follow-up issue to the broad #29

[hfrick]

For this we'd need to subset the model object - and then set it back to the original value. Even when we modify what would usually be a copy.

library(mboost)
#> Loading required package: parallel
#> Loading required package: stabs
library(survival)

lung2 <- lung[-14, ]

f_fit <- blackboost(
  Surv(time, status) ~ age + ph.ecog,
  data = lung2,
  family = CoxPH()
)

# prints with mstop = 100 boosting iterations
f_fit
#> 
#>   Model-based Boosting
#> 
#> Call:
#> blackboost(formula = Surv(time, status) ~ age + ph.ecog, data = lung2,     family = CoxPH())
#> 
#> 
#>   Cox Partial Likelihood 
#> 
#> Loss function:  
#> 
#> Number of boosting iterations: mstop = 100 
#> Step size:  0.1 
#> Offset:  0 
#> Number of baselearners:  1

# `survFit()` does not take an argument for the number of boosting iterations
pred100 <- survFit(f_fit, newdata = lung2)

# change the number of boosting iterations to 10
f_fit[10]
#> 
#>   Model-based Boosting
#> 
#> Call:
#> blackboost(formula = Surv(time, status) ~ age + ph.ecog, data = lung2,     family = CoxPH())
#> 
#> 
#>   Cox Partial Likelihood 
#> 
#> Loss function:  
#> 
#> Number of boosting iterations: mstop = 10 
#> Step size:  0.1 
#> Offset:  0 
#> Number of baselearners:  1

# this has changed `f_fit`
f_fit
#> 
#>   Model-based Boosting
#> 
#> Call:
#> blackboost(formula = Surv(time, status) ~ age + ph.ecog, data = lung2,     family = CoxPH())
#> 
#> 
#>   Cox Partial Likelihood 
#> 
#> Loss function:  
#> 
#> Number of boosting iterations: mstop = 10 
#> Step size:  0.1 
#> Offset:  0 
#> Number of baselearners:  1

# gives you different predictions
pred10 <- survFit(f_fit, newdata = lung2)
identical(pred100, pred10)
#> [1] FALSE

# change it back to 100
f_fit[100]
#> 
#>   Model-based Boosting
#> 
#> Call:
#> blackboost(formula = Surv(time, status) ~ age + ph.ecog, data = lung2,     family = CoxPH())
#> 
#> 
#>   Cox Partial Likelihood 
#> 
#> Loss function:  
#> 
#> Number of boosting iterations: mstop = 100 
#> Step size:  0.1 
#> Offset:  0 
#> Number of baselearners:  1

pred100_2 <- survFit(f_fit, newdata = lung2)
identical(pred100, pred100_2)
#> [1] TRUE

# alternative for changing mstop
mstop(f_fit) <- 50
f_fit
#> 
#>   Model-based Boosting
#> 
#> Call:
#> blackboost(formula = Surv(time, status) ~ age + ph.ecog, data = lung2,     family = CoxPH())
#> 
#> 
#>   Cox Partial Likelihood 
#> 
#> Loss function:  
#> 
#> Number of boosting iterations: mstop = 50 
#> Step size:  0.1 
#> Offset:  0 
#> Number of baselearners:  1

# also simply let's a query the current value
mstop(f_fit)
#> [1] 50

# also note this behavior
fit_50  <- f_fit
mstop(f_fit) <- 70
# this is also 70 now, even though we assigned it to fit_50 before changing mstop for f_fit
mstop(fit_50)
#> [1] 70

^{Created on 2026-05-27 with reprex v2.1.1}