Fit a pure-R isolation forest on rCLR log-ratio inputs — fit_isolation_forest

Builds a pure-R isolation forest (Liu et al., 2008) on rCLR-transformed compositional inputs. Anomalies have shorter expected path lengths to isolation-tree leaves. The decision threshold is calibrated on training scores to control the empirical false-positive rate at fpr_target (applied at scoring time in apply_isolation_forest_logratio).

This implementation uses no external dependencies and is intended for robustness and auditability. For production use on large panels, a C++ backend (e.g., isotree) would be faster.

Usage

fit_isolation_forest_logratio(
  x_train,
  n_trees = 100L,
  sample_size = 256L,
  max_depth = NULL,
  pseudocount = 0.5,
  seed = 42L
)

Arguments

x_train: Numeric matrix (samples \(\times\) features).
n_trees: Integer - number of isolation trees. Default 100.
sample_size: Integer - subsample size per tree. Default 256.
max_depth: Integer or NULL. Maximum tree depth. Default ceiling(log2(min(sample_size, n))).
pseudocount: Numeric. rCLR pseudocount. Default 0.5.
seed: Integer - RNG seed. Default 42.

Value

Object of class isolation_forest_logratio_fit.

References

Liu FT, Ting KM, Zhou Z-H. (2008) Isolation Forest. ICDM.

Examples

if (FALSE) { # \dontrun{
set.seed(42)
x_train <- matrix(abs(rnorm(80 * 20, 100, 30)), nrow = 80,
                  dimnames = list(NULL, paste0("miR-", seq_len(20))))
fit <- fit_isolation_forest_logratio(x_train, n_trees = 20L, sample_size = 32L)
} # }