TuiML Documentation

IQROutlierDetector transformer.

IQR-based outlier detection and handling.

Classes

IQROutlierDetector

class preprocessing.outliers.iqr.IQROutlierDetector(Transformer)

Detect and handle outliers using the Interquartile Range (IQR).

Identifies "extreme" values that fall far outside the central 50% of the data distribution.

Constructor

__init__(
    self,
    factor: float = 1.5,
    action: str = 'clip',
    columns: Optional[List[int]] = None,
)

Overview

The IQR method is a non-parametric outlier detection technique. It defines a "normal" range based on the distance between the first and third quartiles.

Theory

The Interquartile Range (IQR) is $IQR = Q3 - Q1$ . A value $x$ is considered an outlier if:

x < Q1 - k \cdot IQR \quad \text{or} \quad x > Q3 + k \cdot IQR

where $k$ is the multiplier (typically 1.5).

Parameters

factor

float = 1.5

The multiplier :math:`k`.

•1.5: Detects "mild" outliers (Tukey's standard).
•3.0: Detects "extreme" outliers.

action

{"clip", "nan", "remove"} = "clip"

Strategy to handle detected outliers:

•"clip": Replace outliers with the nearest boundary value.
•"nan": Replace outliers with np.nan.
•"remove": Delete rows containing outliers (use with caution in pipelines).

columns

list of int

Indices of columns to process. If None, all columns are checked.

Attributes

bounds_

dict

Mapping of column index to the calculated (lower, upper) boundaries.

Notes

Robustness:

Since it uses quartiles, this method is less sensitive to outliers than

mean/sigma-based methods.

It assumes a unimodal distribution but not necessarily normality.

Remove outliers from a distribution:

python

>>> from tuiml.preprocessing.outliers import IQROutlierDetector
>>> import numpy as np
>>> X = np.array([[10], [12], [11], [10.5], [100.0]])
>>> detector = IQROutlierDetector(action="clip")
>>> X_clean = detector.fit_transform(X)

Methods

get_parameter_schema (cls)

fit (self, X: np.ndarray, y: Optional[np.ndarray]=None, feature_names: Optional[List[str]]=None) -> 'IQROutlierDetector'

transform (self, X: np.ndarray) -> np.ndarray

bounds_ (self)

__repr__ (self) -> str

iqr.py