pp_fck_static — Static fractional convolutional-kernel bank

Group: Feature extraction · C ABI: n4m_transform_fck_static_*

Description

Static fractional convolutional kernel bank transformer.

Parameters

Name

Type

Default

kernel_size

int

—

alphas

Sequence[float] | None

None

sigmas

Sequence[float] | None

None

API and bindings

C ABI (ABI 2): n4m_transform_fck_static_create · n4m_transform_fck_static_destroy · n4m_transform_fck_static_output_cols · n4m_transform_fck_static_transform. Use the linked public header for the exact signature, configuration, and result handles.

Python (verified public re-export):

from n4m.transform.specialized import FCKStaticTransformer

Source signature: FCKStaticTransformer(kernel_size: int, alphas: Sequence[float] | None = None, sigmas: Sequence[float] | None = None, *, filter_orders: Sequence[float] | None = None, filter_scales: Sequence[float] | None = None).

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 canonical paper uniquely defines this n4m operator. It is an implementation-specific bank of fractional convolutional kernels; its generated kernel equation and fixed \(\sigma=3\) are defined by fck_kernel.h and fck_static.c.

Mathematical principle

At construction, one length-kernel_size kernel is generated for every Cartesian pair of fractional order alpha and scale. Each row is convolved with all kernels using nearest-edge extension, and the bands are concatenated into n_kernels * n_features outputs.

Appropriate uses

Fixed multiscale, fractional-order feature expansion before a linear or sparse model.

Limits and validation

Output dimensionality grows multiplicatively with the kernel bank. Edge clamping and the fixed kernel parameterization may not suit every wavelength grid; it is not a learned convolutional model.

Implementation

n4m.transform.specialized.FCKStaticTransformer calls n4m_transform_fck_static_*; kernel construction and convolution are in preprocessing/specialized/fck_static.c.

The ABI-2 implementation is the n4m_transform_fck_static_* lifecycle in libn4m.

Sources and provenance

https://github.com/GBeurier/nirs4all-methods/blob/main/cpp/src/core/preprocessing/specialized/fck_static.c; https://github.com/GBeurier/nirs4all-methods/blob/main/cpp/src/core/common/fck_kernel.h


See also: methods index · interactive dashboard