culebra.solution.feature_selection module

Species and solutions for feature selection problems.

This module provides all the classes necessary to solve feature selection problems with culebra. The possible solutions to the problem are handled by:

  • A Solution abstract class defining the interface for feature selectors.

  • A Species class to define the characteristics that the desired selected features should meet.

Two ndarray-based implementations of the Solution class are also provided:

  • BinarySolution: Solutions are coded as boolean vectors of a fixed length, the number of selectable features. One boolean value is associated to each selectable feature. If the boolean value is True, then the feature is selected, otherwise the feature is ignored.

  • IntSolution: Solutions are coded as vectors of feature indices. Only features whose indices are present in the solution are selected. Thus, solutions may have different lengths and can not contain repeated indices.

In order to make possible the application of evolutionary approaches to this problem, the following Individual implementations are provided:

Ant Colony Optimization is also supported by the Ant class, which inherits from IntSolution and culebra.solution.abc.Ant to add all the path handling stuff to feature index-based solutions.

Finally, this module also provides the Metrics class, which supplies several metrics about the selected features finally obtained.

Attributes

DEFAULT_PROP = 0.15

Default proportion for the generation of a species

MAX_PROP = 0.25

Maximum proportion for the generation of a species