`filter_x_outlier` — Multistrategy X-space outlier filter¶

Group: Sample / feature filters · C ABI: n4m_outlier_detection_x_outlier_*

Description¶

Multivariate outlier filter on the design matrix X.

Full binding docstring

Multivariate outlier filter on the design matrix ``X``.

``method`` selects one of six scoring strategies; see
:c:type:`n4m_filter_x_outlier_method_t`.

Parameters¶

Name	Type	Default
`method`	`str`	`'mahalanobis'`
`use_threshold`	`bool`	`False`
`threshold`	`float`	`0.0`
`n_components`	`int`	`0`
`contamination`	`float`	`0.1`
`seed`	`int`	`0`
`n_estimators`	`int`	`100`
`max_samples`	`int`	`256`

API and bindings¶

C ABI (ABI 2): n4m_outlier_detection_x_outlier_apply · n4m_outlier_detection_x_outlier_create · n4m_outlier_detection_x_outlier_destroy · n4m_outlier_detection_x_outlier_fit · n4m_outlier_detection_x_outlier_is_fitted. Use the linked public header for the exact signature, configuration, and result handles.

Python (verified public re-export):

from n4m.outlier_detection import XOutlierFilter

Source signature: XOutlierFilter(method: str = 'mahalanobis', use_threshold: bool = False, threshold: float = 0.0, n_components: int = 0, contamination: float = 0.1, seed: int = 0, n_estimators: int = 100, max_samples: int = 256).

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¶

The facade combines distinct methods: Liu, Ting & Zhou (2008), Isolation Forest, DOI 10.1109/ICDM.2008.17 (https://doi.org/10.1109/ICDM.2008.17); Breunig et al. (2000), LOF, DOI 10.1145/342009.335388 (https://doi.org/10.1145/342009.335388); and Rousseeuw & Van Driessen (1999), FAST-MCD, DOI 10.1080/00401706.1999.10485670 (https://doi.org/10.1080/00401706.1999.10485670). Mahalanobis and PCA Q/T² options are classical.

Mathematical principle¶

method selects empirical Mahalanobis distance, simplified FAST-MCD robust Mahalanobis, PCA residual \(Q_i=\|x_i-\hat x_i\|^2\), PCA-score Hotelling-like \(T_i^2\), isolation-forest mean path score, or LOF with \(k=\min(20,n-1)\). Defaults use a chi-square-derived distance cutoff for Mahalanobis, training 95th percentiles for PCA scores, and the contamination quantile for forest/LOF.

Appropriate uses¶

Choosing a global, robust, subspace, isolation, or local-density detector under one mask API.

Limits and validation¶

The native robust covariance, isolation forest, and LOF are independent vendored implementations, not calls to scikit-learn and not guaranteed numerically identical. Distance methods are sensitive to scaling; contamination fixes an expected fraction rather than an absolute scientific criterion.

Implementation¶

Python role API n4m.outlier_detection.XOutlierFilter; ABI 2 family n4m_outlier_detection_x_outlier_{create,fit,apply,destroy}; vendored kernels live under cpp/src/core/filters/_vendored/.

The ABI-2 implementation is the n4m_outlier_detection_x_outlier_* lifecycle in libn4m.

Sources and provenance¶

https://github.com/GBeurier/nirs4all-methods/blob/main/cpp/src/core/filters/x_outlier.h; https://doi.org/10.1109/ICDM.2008.17; https://doi.org/10.1145/342009.335388; https://doi.org/10.1080/00401706.1999.10485670

See also: methods index · interactive dashboard

filter_x_outlier — Multistrategy X-space outlier filter¶