Fit random forest model and create model lists

Usage

fit_forest(
  X,
  Y = NULL,
  type = c("regression", "classification", "unsupervised"),
  nodedepth = NULL,
  max_depth = NULL,
  nodesize = NULL,
  ntree = 200,
  forest.wt = "all",
  proximity = "all",
  mtry = NULL,
  ytry = NULL,
  nsplit = 10,
  samptype = c("swor", "swr"),
  nthread = getOption("multiRF.nthread", 0L),
  xvar.wt = NULL,
  yvar.wt = NULL,
  split.wt = NULL,
  case.wt = NULL,
  seed = 529L,
  engine = getOption("multiRF.engine", "native"),
  enhanced_prox = FALSE,
  sibling_gamma = 0.5,
  leaf_embed_dim = 10L,
  ...
)

fit_multi_forest(
  dat.list,
  connect_list = NULL,
  var.wt = NULL,
  yprob = 1,
  ytry = NULL,
  seed = 529L,
  parallel_connections = FALSE,
  cores_connections = NULL,
  ...
)

fit_rfsrc(
  X,
  Y = NULL,
  type = "regression",
  nodedepth = NULL,
  max_depth = NULL,
  nodesize = NULL,
  ntree = 200,
  forest.wt = "all",
  proximity = "all",
  mtry = NULL,
  ytry = NULL,
  nsplit = 10,
  samptype = c("swor", "swr"),
  nthread = getOption("multiRF.nthread", 0L),
  xvar.wt = NULL,
  yvar.wt = NULL,
  split.wt = NULL,
  case.wt = NULL,
  seed = 529L,
  engine = getOption("multiRF.engine", "native"),
  enhanced_prox = FALSE,
  sibling_gamma = 0.5,
  leaf_embed_dim = 10L,
  ...
)

fit_multi_rfsrc(
  dat.list,
  connect_list = NULL,
  var.wt = NULL,
  yprob = 1,
  ytry = NULL,
  seed = 529L,
  ...
)

Arguments

X: A data frame that consider as predictor set
Y: A data frame that consider as response set. If Y = NULL, an unsupervised RF is conducted
type: Select the type of RF model. The default is regression. Can select from "regression", "classification", and "unsupervised"
nodedepth: Backward-compatible alias for max_depth.
max_depth: Maximum tree depth; 0 means unlimited.
nodesize: Minimum terminal-node size. Defaults to 5 for regression, 1 for classification, and 3 for unsupervised forests.
ntree: Number of trees.
forest.wt: Forest-weight output mode: "all" gives all-sample terminal-node weights, "inbag" restricts donors to bootstrap samples, and "oob" evaluates each target only on trees where it is OOB.
proximity: Proximity output mode: "all", "inbag", "oob", or "none".
mtry: Number of candidate X variables per split. Default NULL = ceiling(px/3) for multivariate regression.
ytry: Number of response variables randomly selected per split. Default NULL uses ceiling(qy/3) in multiRF. Set to a specific integer to override (e.g., ytry = ncol(Y) / 2).
nsplit: Number of candidate numeric cutpoints evaluated per variable. multiRF and randomForestSRC both default to 10; set 0 to scan all.
samptype: Sampling scheme: "swor" (without replacement) or "swr" (with replacement).
nthread: Number of threads used for fitting.
xvar.wt, yvar.wt: Optional non-negative predictor/response sampling weights used by the multiRF multivariate forest.
split.wt, case.wt: Optional split/case weights. These are forwarded to randomForestSRC; the default multiRF implementation fails explicitly because it does not yet implement them.
seed: Random seed passed to the selected backend.
engine: Forest backend. Default is getOption("multiRF.engine", "native"). The value "native" selects the default multiRF implementation. randomForestSRC is used only as an optional fallback when explicitly requested.
enhanced_prox: Logical; whether to compute enhanced proximity during multiRF fitting.
sibling_gamma: Strength of the sibling-leaf correction used by enhanced proximity.
leaf_embed_dim: Embedding dimension used by the multiRF enhanced proximity path.
...: Additional arguments passed to randomForestSRC::rfsrc() when engine != "native".
dat.list: A list that contains multi-omics datasets with samples in rows and features in columns. Samples should be matched in each dataset.
connect_list: A pre-defined connection list between datasets. If NULL, all directed pairwise connections are enumerated.
var.wt: Optional named list of variable weights; names must cover the block names in names(dat.list).
yprob: Deprecated. Use ytry directly instead.
parallel_connections: Logical; whether directed connections are fitted in fresh PSOCK worker processes.
cores_connections: Optional core budget for connection-level parallelism.

Value

A model list

Details

fit_forest() uses multiRF by default for classification, multivariate regression, and unsupervised fitting. randomForestSRC is optional and is only used when engine != "native". If type is omitted and Y = NULL, unsupervised fitting is selected.