Row-wise robust standardisation

Centres each row of a matrix by its median and scales by its median absolute deviation (MAD) — a robust alternative to a z-score. Rows with zero MAD (constant values) are returned centred (all zero) instead of NaN.

Usage

standardize(z, na.rm = TRUE)

Arguments

z

A numeric matrix (features x samples).

na.rm

Whether to ignore NAs when computing the median and MAD.

Value

A matrix the same shape as z, robustly standardised by row.

Examples

m <- matrix(c(1, 2, 3, 10, 20, 30), nrow = 2, byrow = TRUE)
standardize(m)
#>            [,1] [,2]      [,3]
#> [1,] -0.6744908    0 0.6744908
#> [2,] -0.6744908    0 0.6744908
# a constant row no longer produces NaN:
standardize(rbind(c(5, 5, 5), c(1, 2, 3)))
#>            [,1] [,2]      [,3]
#> [1,]  0.0000000    0 0.0000000
#> [2,] -0.6744908    0 0.6744908