Skip to contents

Compare and Diff Outer-Fold Runs Across Platforms With Tolerance

Source: R/compare_censoring_weights.R
compare_runs.Rd

Evaluates reproducibility between two sets of model evaluation runs (e.g. from [score_outer()]), such as runs performed across different operating systems (Windows, Linux, macOS) or hardware architectures (x86_64 vs ARM64). Compares integrated Brier scores (IBS), selected penalties (\(\lambda\)), and coefficient estimates within user-specified numeric tolerances.

Usage

compare_runs(
  run1,
  run2,
  tolerance = 1e-06,
  coef_tolerance = tolerance,
  label1 = "run1",
  label2 = "run2"
)

Arguments

run1

A data frame returned by [score_outer()] or a character path to an `.rds` file.

run2

A second data frame or character path to compare against `run1`.

tolerance

Numeric tolerance for IBS and \(\lambda\) differences (default `1e-6`).

coef_tolerance

Numeric tolerance for coefficient differences (defaults to `tolerance`).

label1

Optional character label for `run1` (default `"run1"`).

label2

Optional character label for `run2` (default `"run2"`).

Value

An object of class `c("templecbe_run_comparison", "tbl_df", "tbl", "data.frame")` containing per-model diffs, with overall reproducibility summary attributes.

See also

[score_outer()], [tidy_outer_coefficients()]