Keeps the predictors whose odds (or hazard) ratio is above a threshold, learns the size and term frequencies of the formulas of the best models, and draws new candidate formulas from them. A front end to [sample_candidate_formulas_adaptive()] for a table of predictor ratios and a table of already selected (low-error) models.
Usage
adaptive_cox_formulas(
summary_df,
low_ib_models,
n_formulas = 5,
or_threshold = 1,
seed = NULL
)Arguments
- summary_df
Data frame with `predictor` and `odds_ratio` (ratios; use `exp(coef)` for Cox models). Predictors are matched to the terms of the formulas by name.
- low_ib_models
Data frame of the best models with a `formula` column (text or formula objects). All rows count as winners; rank and filter them first.
- n_formulas
Number of distinct formulas wanted.
- or_threshold
Keep predictors with a ratio **above** this value (1 = harmful or associated with more events).
- seed
Optional integer; see [sample_candidate_formulas()].
Examples
ratios <- data.frame(predictor = c("a", "b", "c", "d"), odds_ratio = c(1.8, 0.6, 1.2, 2.5))
top <- data.frame(formula = c("y ~ a + c", "y ~ a + d", "y ~ a + c + d"))
adaptive_cox_formulas(ratios, top, n_formulas = 3, seed = 1)
#> # A tibble: 3 × 4
#> id formula terms n_terms
#> <int> <list> <list> <int>
#> 1 1 <formula> <chr [2]> 2
#> 2 2 <formula> <chr [2]> 2
#> 3 3 <formula> <chr [2]> 2