filter_correlation — Absolute Pearson-correlation feature filter¶
Group: Signal transforms · C ABI: n4m_feature_selection_correlation_*
Description¶
Model-agnostic feature filter by absolute correlation to y.
Parameters¶
Name |
Type |
Default |
|---|---|---|
|
|
|
|
|
|
API and bindings¶
C ABI (ABI 2): n4m_feature_selection_correlation_create · n4m_feature_selection_correlation_destroy · n4m_feature_selection_correlation_fit · n4m_feature_selection_correlation_is_fitted · n4m_feature_selection_correlation_output_cols · n4m_feature_selection_correlation_transform. Use the linked public header for the exact signature, configuration, and result handles.
Python (verified public re-export):
from n4m.feature_selection.filter import CorrelationFilter
Source signature: CorrelationFilter().
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¶
There is no single canonical feature-selection paper for this use of Pearson correlation. The absolute-score threshold and top-k policy are implementation-specific.
Mathematical principle¶
For each channel, compute the centered Pearson coefficient \(r_j=\sum_i(x_{ij}-\bar x_j)(y_i-\bar y)/[\sqrt{\sum_i(x_{ij}-\bar x_j)^2}\sqrt{\sum_i(y_i-\bar y)^2}]\) and score it by \(|r_j|\). Zero-denominator features score zero. Retain scores strictly above threshold or the top_k largest.
Appropriate uses¶
Fast supervised screening of wavelengths with a strong marginal linear relation to one target.
Limits and validation¶
It is univariate, linear, and ignores redundancy between selected wavelengths. Absolute value removes sign information; constant y makes every score zero; using it before cross-validation would leak targets.
Implementation¶
Python role API n4m.feature_selection.filter.CorrelationFilter; ABI 2 family n4m_feature_selection_correlation_{create,fit,transform,output_cols,destroy}.
The ABI-2 implementation is the n4m_feature_selection_correlation_* lifecycle in libn4m.
Sources and provenance¶
https://github.com/GBeurier/nirs4all-methods/blob/main/cpp/src/c_api/c_api_advanced.cpp#L1138
See also: methods index · interactive dashboard