ws_mad_logratio.Rd

% Generated by roxygen2: do not edit by hand
% Please edit documentation in R/singlesample-within-sample.R
\name{ws_mad_logratio}
\alias{ws_mad_logratio}
\title{Median-centred log-ratio with optional MAD scaling}
\usage{
ws_mad_logratio(x, pseudocount = 0.5, scale_by_mad = TRUE)
}
\arguments{
\item{x}{Samples x features matrix (rows = samples).}

\item{pseudocount}{Additive pseudocount before logging. Default 0.5.}

\item{scale_by_mad}{Logical. If \code{TRUE}, divides each sample by its
per-sample MAD after median-centering. Default \code{TRUE}.}
}
\value{
Same shape as \code{x}; the median-centred (MAD-scaled)
  log-abundance values, with attributes \code{centering} and
  \code{scaling}.
}
\description{
Robust descriptive transform on log-abundances: subtracts the per-sample
median log-abundance (more robust to outlier features than mean-centering
as in CLR/rCLR) and optionally divides by the per-sample median absolute
deviation (MAD; limits leverage from individual high-magnitude features).

This is NOT an Aitchison-valid simplex projection. It is a robust
descriptive transform on log-abundances, suitable as input to learners
that do not require the zero-sum compositional coordinate constraint.
}