pp_wavelet_svd — Truncated SVD of multilevel wavelet coefficients¶
Group: Augmentation · C ABI: n4m_transform_wavelet_svd_*
Description¶
DWT coefficient projection through SVD scores.
Parameters¶
Name |
Type |
Default |
|---|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
API and bindings¶
C ABI (ABI 2): n4m_transform_wavelet_svd_create · n4m_transform_wavelet_svd_destroy · n4m_transform_wavelet_svd_fit · n4m_transform_wavelet_svd_is_fitted · n4m_transform_wavelet_svd_output_cols · n4m_transform_wavelet_svd_transform. Use the linked public header for the exact signature, configuration, and result handles.
Python (verified public re-export):
from n4m.transform.wavelet import WaveletSVD
Source signature: WaveletSVD(family: str = 'haar', mode: str = 'periodization', max_level: int = 2, n_components: float = 5.0).
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¶
No unique paper defines this composite. It combines Mallat’s DWT (https://doi.org/10.1109/34.192463) with Eckart–Young low-rank approximation (https://doi.org/10.1007/BF02288367).
Mathematical principle¶
Rows are mapped to packed multilevel DWT coefficients and compact SVD is fitted without the explicit centering used by WaveletPCA. The operator keeps an integer rank or the smallest rank reaching the requested singular-value energy fraction and emits right-singular-vector scores.
Appropriate uses¶
Low-rank multiscale features when retaining the coefficient-matrix mean direction is intentional.
Limits and validation¶
Uncentered SVD can spend its first direction on mean offset. The learned basis is training-dependent, shift-sensitive through the DWT, and output rank may vary for a fraction threshold.
Implementation¶
n4m.transform.wavelet.WaveletSVD calls n4m_transform_wavelet_svd_*; coefficient packing and uncentered SVD are in wavelet_svd.c.
The ABI-2 implementation is the n4m_transform_wavelet_svd_* lifecycle in libn4m.
Sources and provenance¶
https://doi.org/10.1109/34.192463; https://doi.org/10.1007/BF02288367; https://github.com/GBeurier/nirs4all-methods/blob/main/cpp/src/core/preprocessing/wavelets/wavelet_svd.c
See also: methods index · interactive dashboard