Random Subset feature selection.
This module provides random feature selection, which randomly selects a subset of features. Useful for ensemble methods and feature bagging.
Classes
class features.selection.random_subset.RandomSubsetSelector(FeatureSelector, SelectorMixin)
Randomly select a subset of features.
Chooses a random subset of features, either an absolute number or a percentage. Useful for ensemble methods, feature bagging, and reducing dimensionality when labels are unavailable.
Constructor
__init__( self, n_features: Union[int, float] = 0.5, invert: bool = False, random_state: Optional[int] = None, )
Overview
This selector samples a set of feature indices without replacement. It is a purely stochastic method and does not use any information from the data values or target labels during selection.
Parameters
n_features
int or float
= 0.5
Number of features to select:
- •If
int >= 1: absolute number of features. - •If
float < 1: fraction of total features.
invert
bool
= False
If True, randomly removes features instead of selecting them (retains all features NOT in the random sample).
random_state
int
Random seed for reproducibility.
Attributes
selected_features_
np.ndarray
Indices of the randomly selected features.
n_features_selected_
int
Total number of features selected.
Notes
When to use:
- For creating diverse ensembles via feature bagging.
- To benchmark other feature selection methods against a random baseline.
- When you want to reduce dimensionality quickly without any statistical assumptions.
- Highly likely to keep irrelevant or redundant features.
- Results depend entirely on the random seed.
See Also
Randomly select half of the features:
python
>>> from tuiml.features.selection import RandomSubsetSelector
>>> import numpy as np
>>> X = np.random.randn(10, 20)
>>> selector = RandomSubsetSelector(n_features=0.5, random_state=42)
>>> X_new = selector.fit_transform(X)
>>> print(X_new.shape[1])
10Methods
fit
(self, X: np.ndarray, y: Optional[np.ndarray]=None) -> 'RandomSubsetSelector'
fit
(self, X: np.ndarray, y: Optional[np.ndarray]=None) -> 'RandomSubsetSelector'
Fit the random subset selector.
Parameters
X
ndarray of shape (n_samples, n_features)
Training data.
y
Ignored
Not used, present for API consistency. RandomSubsetSelector is unsupervised.
Returns
self
RandomSubsetSelector
The fitted selector.
class features.selection.random_subset.BootstrapFeaturesSelector(FeatureSelector, SelectorMixin)
Bootstrap feature selection for ensemble methods.
Selects features using bootstrap sampling (sampling with replacement to create a sample of the same size, then taking unique indices), commonly used in Random Forest and other bagging-based ensembles.
Constructor
__init__( self, n_features: Union[int, float, str] = 'sqrt', random_state: Optional[int] = None, )
Overview
Bootstrap feature selection introduces diversity by allowing some features to be sampled multiple times while others are omitted in a single draw. This increases the randomness and robustness of ensemble models.
Parameters
n_features
int, float, or {"sqrt", "log2"}
= "sqrt"
Number of features to draw in the bootstrap sample:
- •
int >= 1: absolute number. - •
float < 1: fraction of total. - •
"sqrt": :math:`\sqrt{n_{features}}`. - •
"log2": :math:`log_2(n_{features})`.
random_state
int
Random seed for reproducibility.
Attributes
selected_features_
np.ndarray
Unique indices selected during the bootstrap process.
n_features_selected_
int
Actual number of unique features selected.
Notes
When to use:
- Specifically for ensemble methods like Random Forest.
- When you want to simulate feature bagging.
Select square root of features using bootstrap:
python
>>> from tuiml.features.selection import BootstrapFeaturesSelector
>>> import numpy as np
>>> X = np.random.randn(10, 100)
>>> selector = BootstrapFeaturesSelector(n_features="sqrt", random_state=42)
>>> X_new = selector.fit_transform(X)
>>> print(X_new.shape[1])
10