prepare_svd_robust_input

Internal helper used by <code>pca_robust()</code> to compute a singular value
decomposition that is less sensitive to individual rows with extreme values.
The routine alternates between computing the SVD of a row-weighted matrix and
updating the weights via a Huber-type scheme based on the reconstruction
residuals.

internal

High performance principal component analysis routines
that operate directly on 'bigmemory::big.matrix' objects. The
package avoids materialising large matrices in memory by
streaming data through 'BLAS' and 'LAPACK' kernels and provides
helpers to derive scores, loadings, correlations, and
contribution diagnostics, including utilities that stream
results into 'bigmemory'-backed matrices for file-based
workflows. Additional interfaces expose 'scalable' singular value
decomposition, robust PCA, and robust SVD algorithms so that
users can explore large matrices while tempering the influence
of outliers. 'Scalable' principal component analysis is also implemented,
Elgamal, Yabandeh, Aboulnaga, Mustafa, and Hefeeda (2015)
<doi:10.1145/2723372.2751520>.

Frederic Bertrand

bigPCAcpp

Principal Component Analysis for 'bigmemory' Matrices

prepare_svd_robust_input function

<dl><dt>x</dt>
<dd>Numeric matrix for which the decomposition should be computed.</dd>
<dt>ncomp</dt>
<dd>Number of leading components to retain.</dd>
<dt>max_iter</dt>
<dd>Maximum number of reweighting iterations.</dd>
<dt>tol</dt>
<dd>Convergence tolerance applied to successive changes in the row
weights and singular values.</dd>
<dt>huber_k</dt>
<dd>Tuning constant controlling the aggressiveness of the Huber
weight function. Larger values down-weight fewer observations.</dd></dl>

Arguments

Prepare iteratively reweighted singular value decomposition — prepare_svd_robust_input

<dl>

<dt>x</dt>
<dd>Numeric matrix for which the decomposition should be computed.</dd>


<dt>ncomp</dt>
<dd>Number of leading components to retain.</dd>


<dt>max_iter</dt>
<dd>Maximum number of reweighting iterations.</dd>


<dt>tol</dt>
<dd>Convergence tolerance applied to successive changes in the row
weights and singular values.</dd>


<dt>huber_k</dt>
<dd>Tuning constant controlling the aggressiveness of the Huber
weight function. Larger values down-weight fewer observations.</dd>

</dl>

prepare_svd_robust_input: Prepare iteratively reweighted singular value decomposition

Description

Usage

Value

Arguments