When using the "mboost" boost_tree for prediction, an error occurs when case weights are applied.

[AKALeon]

library(tidymodels)
library(mboost)
library(censored)

table(lung$status)

> table(lung$status)

  1   2 
 63 165 

data = lung |> 
  mutate(outcome = Surv(time, status == 2),
         case_wts = if_else(status == 2, 1, 165 / 63),
         case_wts = importance_weights(case_wts)) |> 
  select(-time, -status)

boost_tree_spec = boost_tree() |> 
  set_engine('mboost') |> 
  set_mode('censored regression')

rec = recipe(formula = outcome ~ ., data = data) |> 
  step_impute_mean(all_numeric_predictors())

When no case weight is added, everything works normally:

workflow() |> 
  add_recipe(rec) |> 
  add_model(boost_tree_spec) |> 
  fit(data = data) |> 
  augment(new_data = data, 
          eval_time = sort(unique(lung$time))) |> 
  concordance_survival(truth = outcome, 
                       estimate = .pred_time)

# A tibble: 1 × 3
  .metric              .estimator .estimate
  <chr>                <chr>          <dbl>
1 concordance_survival standard       0.736

When case weight is added, there is error:

workflow() |> 
  add_recipe(rec) |> 
  add_model(boost_tree_spec) |> 
  add_case_weights(col = case_wts) |> 
  fit(data = data) |>
  augment(new_data = data, 
          eval_time = sort(unique(lung$time))) |> 
  concordance_survival(truth = outcome, 
                       estimate = .pred_time)

Error in survFit.mboost(object$fit, new_data) : 
  survFit cannot (yet) deal with weights

This error occurs in augment() not in fit():

workflow() |> 
  add_recipe(rec) |> 
  add_model(boost_tree_spec) |> 
  add_case_weights(col = case_wts) |> 
  fit(data = data)

══ Workflow [trained] ═════════════════════════════════════════════════════════════════════
Preprocessor: Recipe
Model: boost_tree()

── Preprocessor ───────────────────────────────────────────────────────────────────────────
1 Recipe Step

• step_impute_mean()

── Case Weights ───────────────────────────────────────────────────────────────────────────
case_wts

── Model ──────────────────────────────────────────────────────────────────────────────────

	 Model-based Boosting

Call:
mboost::blackboost(formula = formula, data = data, weights = weights,     family = family, control = mboost::boost_control(), tree_controls = partykit::ctree_control(teststat = "quadratic",         testtype = "Teststatistic", mincriterion = 0, minsplit = 10,         minbucket = 4, maxdepth = 2, saveinfo = FALSE))


	 Cox Partial Likelihood 

Loss function:  

Number of boosting iterations: mstop = 100 
Step size:  0.1 
Offset:  0 
Number of baselearners:  1 

workflow() |> 
  add_recipe(rec) |> 
  add_model(boost_tree_spec) |> 
  add_case_weights(col = case_wts) |> 
  fit(data = data) |>
  augment(new_data = data, 
          eval_time = sort(unique(lung$time)))

Error in survFit.mboost(object$fit, new_data) : 
  survFit cannot (yet) deal with weights

[hfrick]

Hm, that error is coming from the mboost package, censored only does the wiring up into the parsnip framework. We could turn of weighted modeling fitting but for weighted prediction to work, it would be best to ask the maintainer to implement it. The "(yet)" would imply that they generally intend to do so.

[AKALeon]

@hfrick Thank you!