diss_mahalanobis

Creates a configuration object for computing Mahalanobis dissimilarity.
Pass the result to <code>dissimilarity()</code> to compute the dissimilarity
matrix.
The Mahalanobis distance is computed by first transforming the data into
Mahalanobis space via a factorization of the inverse covariance matrix
M^-1 = W^TWM^-1 = W'W (using SVD), then applying Euclidean distance in
that transformed space:
d(x_i, x_j) = 1p(x_i - x_j)M^-1(x_i - x_j)^Td(x_i, x_j) = sqrt(1/p * (x_i - x_j) M^-1 (x_i - x_j)^T)

Functions for dissimilarity analysis and machine learning in complex
spectral data sets, including memory-based learning (MBL), optimal subset
search and selection, and retrieval-based modelling with model libraries.
Supports local learning, optimisation of spectral libraries, and
ensemble prediction from precomputed models. Most of these functions are
based on the methods presented in Ramirez-Lopez et al. (2013)
<doi:10.1016/j.geoderma.2012.12.014>, Ramirez-Lopez et al. (2026a)
<doi:10.1016/j.aca.2026.345682>, and Ramirez-Lopez et al. (2026b)
<doi:10.1016/j.aca.2026.345651>.

Leonardo RamirezLopez

resemble

Similarity Retrieval and Local Learning for Spectral
Chemometrics

Leonardo Ramirez-Lopez

Antoine Stevens

Claudio Orellano

diss_mahalanobis function

<dl><dt>center</dt>
<dd>Logical. Center the data before computing distances?
Applied jointly to <code>Xr</code> and <code>Xu</code> if both are provided.
Default <code>TRUE</code>.</dd>
<dt>scale</dt>
<dd>Logical. Scale the data before computing distances?
Applied jointly to <code>Xr</code> and <code>Xu</code> if both are provided.
Default <code>FALSE</code>.</dd></dl>

Arguments

The covariance matrix will be singular — and the distance therefore
uncomputable — when the number of observations is smaller than the number
of variables, or when variables are perfectly collinear. This is common
with raw spectral data; consider using <code>diss_euclidean()</code> on
PCA scores instead.

Important limitations

<a href="https://orcid.org/0000-0002-5369-5120">Leonardo Ramirez-Lopez</a>

Author

Mahalanobis dissimilarity method constructor — diss_mahalanobis

<dl>

<dt>center</dt>
<dd>Logical. Center the data before computing distances?
Applied jointly to <code>Xr</code> and <code>Xu</code> if both are provided.
Default <code>TRUE</code>.</dd>


<dt>scale</dt>
<dd>Logical. Scale the data before computing distances?
Applied jointly to <code>Xr</code> and <code>Xu</code> if both are provided.
Default <code>FALSE</code>.</dd>

</dl>

The covariance matrix will be singular — and the distance therefore
uncomputable — when the number of observations is smaller than the number
of variables, or when variables are perfectly collinear. This is common
with raw spectral data; consider using <code>diss_euclidean()</code> on
PCA scores instead.

<a href='https://orcid.org/0000-0002-5369-5120'>Leonardo Ramirez-Lopez</a>

diss_mahalanobis: Mahalanobis dissimilarity method constructor

Description

Usage

Value

Arguments

Important limitations

Author

See Also

Examples