Skip to contents

mrf3 Stability Evaluation (No Refit)

Source: R/workflow_summary_tools.R
mrf3_stability.Rd

Evaluate clustering stability by repeated subsampling/bootstrapping on existing mrf3_fit outputs, without refitting forests.

Usage

mrf3_stability(
  x,
  branches = c("specific_shared", "robust_clustering"),
  n_rep = 50L,
  sample_frac = 0.8,
  sample_mode = c("subsample", "bootstrap"),
  specific_k = NULL,
  robust_k = NULL,
  specific_method = c("auto", "PAM", "Spectral"),
  robust_method = c("auto", "PAM", "Spectral"),
  run_consensus = TRUE,
  consensus_method = c("PAM", "Spectral"),
  truth = NULL,
  seed = 529,
  verbose = TRUE
)

Arguments

x

An object from mrf3_fit().

branches

Branches to evaluate. Supported: "specific_shared" and "robust_clustering". Branches unavailable in x are skipped with a warning; an error is raised only if none of the requested branches are available.

n_rep

Number of repeats.

sample_frac

Sampling fraction per repeat.

sample_mode

Sampling mode: "subsample" or "bootstrap".

specific_k

Optional k override for "specific_shared" branch.

robust_k

Optional k override for "robust_clustering" branch.

specific_method

Clustering backend for "specific_shared" repeats: "auto", "Spectral", or "PAM".

robust_method

Clustering backend for "robust_clustering" repeats: "auto", "Spectral", or "PAM".

run_consensus

Logical; whether to derive consensus clustering.

consensus_method

Consensus clustering backend: "Spectral" or "PAM".

truth

Optional truth labels for external metrics. If names are provided, they are aligned to sample names.

seed

Random seed.

verbose

Logical; whether to print progress.

Value

A list with per-branch repeat metrics and optional consensus results.