API Reference / features / selection /

rfe.py

Recursive feature elimination selectors.

Classes

RFE

class features.selection.rfe.RFE(FeatureSelector, SelectorMixin)

Recursive feature elimination.

Constructor 
__init__(
    self,
    estimator: Any = None,
    n_features_to_select: int | None = None,
    step: float = 1,
)

Parameters

estimator
object
A supervised estimator exposing fit and either feature_importances_ or coef_ after fitting.
n_features_to_select
int
Number of features to retain. If None, half of the input features are retained.
step
int or float = 1
Number of features to remove at each iteration, or the fraction of features to remove when a float is supplied.

Attributes

n_features_to_select_
int
Number of features selected after fitting.
support_
ndarray of bool
Boolean mask indicating selected features.
ranking_
ndarray of int
Feature ranking. Selected features have rank 1.
python 
>>> import numpy as np
>>> from tuiml.features.selection import RFE
>>> from tuiml.algorithms.trees.random_forest import RandomForestClassifier
>>> X = np.random.RandomState(0).randn(30, 5)
>>> y = np.random.RandomState(0).randint(0, 2, 30)
>>> selector = RFE(
...     estimator=RandomForestClassifier(n_estimators=5, random_state=0),
...     n_features_to_select=2,
... )
>>> _ = selector.fit(X, y)
>>> int(selector.support_.sum())
2

Methods

fit (self, X: np.ndarray, y: np.ndarray | None=None) -> 'RFE'

Fit RFE and recursively eliminate features.

Parameters
X
ndarray of shape (n_samples, n_features)
Training feature matrix.
y
ndarray of shape (n_samples,)
Target values.
Returns
self
RFE
The fitted selector.
get_parameter_schema (cls) -> dict

Return the parameter schema for RFE.

RFECV

class features.selection.rfe.RFECV(RFE)

Recursive feature elimination with cross-validation.

Constructor 
__init__(
    self,
    estimator: Any = None,
    step: float = 1,
    cv: Any = 5,
    scoring: Any | None = None,
    random_state: int | None = None,
)

Parameters

estimator
object
A supervised estimator exposing fit, predict and either feature_importances_ or coef_ after fitting.
step
int or float = 1
Number of features to remove at each iteration, or the fraction of features to remove when a float is supplied.
cv
int or splitter = 5
Number of folds when an integer is supplied, or a splitter exposing split(X, y).
scoring
str or callable
Scoring strategy. If None, classification uses accuracy and regression uses R².
random_state
int
Random seed used when cv is an integer.

Attributes

n_features_to_select_
int
Number of features selected by cross-validation.
support_
ndarray of bool
Boolean mask indicating selected features.
ranking_
ndarray of int
Feature ranking.
cv_results_
dict
Cross-validation results for each tested feature count.
python 
>>> import numpy as np
>>> from tuiml.features.selection import RFECV
>>> from tuiml.algorithms.trees.random_forest import RandomForestClassifier
>>> X = np.random.RandomState(0).randn(40, 5)
>>> y = np.random.RandomState(1).randint(0, 2, 40)
>>> selector = RFECV(
...     estimator=RandomForestClassifier(n_estimators=5, random_state=0),
...     cv=3,
... )
>>> _ = selector.fit(X, y)
>>> selector.n_features_to_select_ >= 1
True

A splitter instance can also be supplied:

python 
>>> from tuiml.evaluation.splitting import KFold
>>> cv = KFold(n_splits=3, shuffle=True, random_state=0)
>>> selector = RFECV(
...     estimator=RandomForestClassifier(n_estimators=5, random_state=0),
...     cv=cv,
... )
>>> _ = selector.fit(X, y)

Methods

fit (self, X: np.ndarray, y: np.ndarray | None=None) -> 'RFECV'

Fit RFECV and select the feature count with the best CV score.

Parameters
X
ndarray of shape (n_samples, n_features)
Training feature matrix.
y
ndarray of shape (n_samples,)
Target values.
Returns
self
RFECV
The fitted selector.
get_parameter_schema (cls) -> dict

Return the parameter schema for RFECV.