nugget_feature_weights: Feature weights derived from data nuggets

In MosaiClusteR: An Umbrella Framework for Multi-Source and Multi-Omics Clustering

Feature weights derived from data nuggets

Description

Computes a per-feature importance score from a data-nugget (create_data_nuggets) object built over the observations (samples). This score is the data-nugget analogue of the feature variance used by the ABC / M-ABC family: it rewards features that separate the representative groups while discounting within-nugget noise, giving a robust, Big-Data-friendly weighting.

Usage

nugget_feature_weights(nuggets, type = c("between", "ratio", "inv_scale"))

Arguments

nuggets	A "data_nugget" object from create_data_nuggets (built with samples in the rows).
type	One of: "between" (default) size-weighted between-nugget variance of each feature: \sum_i w_i (c_{ij} - \bar c_j)^2 / \sum_i w_i. "ratio" a separability / pseudo-F score, \mathrm{between}_j / (\mathrm{between}_j + \overline{\mathrm{within}}_j). "inv_scale" inverse mean within-nugget variance, which down-weights noisy features.

Value

A named numeric vector of length ncol(centers) of non-negative feature scores.

References

\insertRef

Cherasia2023MosaiClusteR

See Also

create_data_nuggets, M_ABC

Examples

set.seed(1)
x <- cbind(signal = c(rnorm(50, 0), rnorm(50, 5)),  # separates groups
           noise  = rnorm(100))                      # pure noise
dn <- create_data_nuggets(x, max_nuggets = 20)
nugget_feature_weights(dn, type = "between")

MosaiClusteR documentation built on Oct. 25, 2026, 1:08 a.m.