pp_flex_svd — Flexible truncated singular-value decomposition¶
Group: Feature extraction · C ABI: n4m_decomposition_flexible_svd_*
Description¶
SVD with integer component selection.
Parameters¶
Name |
Type |
Default |
|---|---|---|
|
|
|
API and bindings¶
C ABI (ABI 2): n4m_decomposition_flexible_svd_create · n4m_decomposition_flexible_svd_destroy · n4m_decomposition_flexible_svd_fit · n4m_decomposition_flexible_svd_is_fitted · n4m_decomposition_flexible_svd_output_cols · n4m_decomposition_flexible_svd_transform. Use the linked public header for the exact signature, configuration, and result handles.
Python (verified public re-export):
from n4m.decomposition import FlexibleSVD
Source signature: FlexibleSVD().
R: no current source-verified entry point was found for this catalog method.
MATLAB / Octave: no current source-verified entry point was found for this catalog method.
Explanations¶
Bibliographic source¶
Eckart & Young (1936), The approximation of one matrix by another of lower rank, Psychometrika 1, 211–218, https://doi.org/10.1007/BF02288367.
Mathematical principle¶
Compact SVD factorizes the uncentered input as \(X=U\Sigma V^T\) and emits the first \(k\) coordinates \(XV_k\). A parameter at least one is truncated to an integer rank; a parameter in \((0,1)\) selects the smallest rank reaching that fraction of total column variance using the projected-component variance ratios.
Appropriate uses¶
Low-rank compression when retaining the mean direction is intentional and rank should be chosen directly or from a variance target.
Limits and validation¶
Uncentered SVD can devote its first direction to mean offset and is scale-sensitive. The fractional criterion is based on projected variance rather than simply squared singular-value energy, and fitting must remain inside validation folds.
Implementation¶
n4m.decomposition.FlexibleSVD calls n4m_decomposition_flexible_svd_*; the compact factorization and transform are in flexible_svd.c.
The ABI-2 implementation is the n4m_decomposition_flexible_svd_* lifecycle in libn4m.
Sources and provenance¶
https://doi.org/10.1007/BF02288367; https://github.com/GBeurier/nirs4all-methods/blob/main/cpp/src/core/preprocessing/feature_selection/flexible_svd.c
See also: methods index · interactive dashboard