Compute Cluster-Specific IMD Without Pairwise IMD

Usage

cluster_imd(
  x,
  cluster = NULL,
  dat.list = NULL,
  connect_list = NULL,
  min_cluster_size = 20L,
  ntree = NULL,
  ytry = NULL,
  parallel = TRUE,
  imd_normalized_weights = FALSE,
  run_vs = FALSE,
  vs_args = list(),
  fit_args = list(),
  imd_args = list(),
  use_node_imd = FALSE,
  keep_model = FALSE,
  keep_data = FALSE,
  seed = 529
)

Arguments

x: A mrf3 or mrf3_fit object.
cluster: Optional cluster labels for samples.
dat.list: Optional named omics list (samples in rows).
connect_list: Optional global connection list used by all clusters.
min_cluster_size: Minimum sample size required to run a cluster-specific IMD.
ntree: Deprecated compatibility argument. Ignored in no-refit mode.
ytry: ytry used for per-cluster IMD. If NULL, reuses object setting. Only used by the depth-based path.
parallel: Logical; whether IMD computation inside each cluster is parallelized. Only used by the depth-based path (the node-score fast path is single-pass and needs no parallelism).
imd_normalized_weights: Logical; passed to get_multi_weights() as normalized. The default is FALSE so cluster-level variable selection receives raw forest IMD on its 0-to-1 scale.
run_vs: Logical; whether to run mrf3_vs() variable selection on each cluster after IMD. run_vs = TRUE forces the depth-based path, which builds the per-cluster subset models that mrf3_vs() requires.
vs_args: A named list of additional arguments passed to mrf3_vs() for each cluster.
fit_args: Deprecated compatibility argument. Ignored in no-refit mode.
imd_args: Optional named list merged into each per-cluster get_multi_weights() call. Supplying any imd_args forces the depth-based path, because the node-score fast path cannot honor get_multi_weights() options.
use_node_imd: Controls the estimator used for the per-cluster weights. FALSE (the default) computes Eq. 6-8 inverse minimal depth via the per-tree network traversal of get_multi_weights() on each cluster's subset model. TRUE uses the node-score fast path (cluster_weighted_imd()): the per-node split statistics stored during fitting are averaged per variable, weighted by the fraction of the cluster's samples whose root-to-leaf paths visit each node. The fast path is deterministic and typically 100-900x faster, but it is a different, split-score-weighted estimator whose weights stay close to the global importance ranking: in benchmarks its per-cluster weights correlate at Spearman 0.9-0.99 across clusters (depth path: 0.6-0.8), and agree with the depth-based weights at Spearman ~0.4-0.6 (top-20 overlap ~50-75%). Use it as a quick descriptive screen, not as a substitute for the depth-based cluster IMD (see cluster_weighted_imd() for the definition). NULL selects the fast path automatically whenever the models carry the required tree_info statistics and no blocking option (imd_args, run_vs) is requested, falling back to the depth-based path otherwise.
keep_model: Logical; whether to keep per-cluster mrf3 objects in output.
keep_data: Logical; whether to keep per-cluster subset data in output.
seed: Base seed; cluster i uses seed + i - 1.

Value

A list with cluster-level summary, per-cluster IMD outputs, and params. params$imd_path records which estimator produced the weights: "node_score" (fast path) or "depth_traversal" (Eq. 6-8 IMD).