`filter_y_outlier` — Univariate target outlier filter¶

Group: Sample / feature filters · C ABI: n4m_outlier_detection_y_outlier_*

Description¶

Univariate outlier filter on the target vector y.

Full binding docstring

Univariate outlier filter on the target vector ``y``.

``method`` is one of ``"iqr"``, ``"zscore"``, ``"percentile"``, ``"mad"``.
Threshold semantics follow nirs4all's :class:`YOutlierFilter`:

* For ``"iqr"`` / ``"zscore"`` / ``"mad"``: ``threshold`` is the
  multiplier on the IQR / σ / MAD bounds.
* For ``"percentile"``: ``lower_percentile`` and ``upper_percentile``
  define the keep band.

Parameters¶

Name	Type	Default
`method`	`str`	`'iqr'`
`threshold`	`float`	`1.5`
`lower_percentile`	`float`	`1.0`
`upper_percentile`	`float`	`99.0`

API and bindings¶

C ABI (ABI 2): n4m_outlier_detection_y_outlier_apply · n4m_outlier_detection_y_outlier_create · n4m_outlier_detection_y_outlier_destroy · n4m_outlier_detection_y_outlier_fit · n4m_outlier_detection_y_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 YOutlierFilter

Source signature: YOutlierFilter(method: str = 'iqr', threshold: float = 1.5, lower_percentile: float = 1.0, upper_percentile: float = 99.0).

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 single paper defines this four-rule facade. IQR fences, z scores, percentiles, and scaled MAD are classical robust/descriptive rules; the exact NumPy-compatible percentile interpolation is source-defined.

Mathematical principle¶

Fit one interval on y: IQR uses \([Q_1-tIQR,Q_3+tIQR]\); z score uses \([\mu-t\sigma,\mu+t\sigma]\); percentile uses configured quantiles; MAD uses \([m-t(1.4826\,MAD),m+t(1.4826\,MAD)]\). Apply keeps values inside the learned bounds and returns mask plus counts. Quantiles use linear interpolation.

Appropriate uses¶

Auditable removal or flagging of extreme reference values before fitting a calibration model.

Limits and validation¶

Filtering y narrows the calibration domain and can bias validation. Z score assumes a meaningful mean/scale; IQR and MAD can degenerate on tied values. Fit bounds only on the training partition.

Implementation¶

Python role API n4m.outlier_detection.YOutlierFilter; ABI 2 family n4m_outlier_detection_y_outlier_{create,fit,apply,is_fitted,destroy}.

The ABI-2 implementation is the n4m_outlier_detection_y_outlier_* lifecycle in libn4m.

Sources and provenance¶

https://github.com/GBeurier/nirs4all-methods/blob/main/cpp/src/core/filters/y_outlier.h

See also: methods index · interactive dashboard

filter_y_outlier — Univariate target outlier filter¶