pyccea.fitness package

class pyccea.fitness.DistanceBasedFitness(evaluator: WrapperEvaluation, weights: list)

Bases: WrapperFitnessFunction

Objective function that maximizes balanced accuracy based on a k-nearest neighbors classifier.

The fitness function is designed as a three-objective optimization, aimed at achieving a balance between maximizing balanced accuracy while simultaneously minimizing the average distance between instances sharing the same label and maximizing the average distance between instances with different labels.

Firouznia, Marjan, Pietro Ruiu, and Giuseppe A. Trunfio. “Adaptive cooperative coevolutionary differential evolution for parallel feature selection in high-dimensional datasets.” The Journal of Supercomputing (2023): 1-30.

Attributes:

w1: float

Predictive performance weight.

w2: float

Weight of the complement of the average distance between instances and their neighbors with the same class.

w3: float

Weight of the average distance between instances and their neighbors of different classes.

Methods

evaluate(context_vector, data)

Evaluate the given context vector using the fitness function.

evaluate(context_vector: ndarray, data: DataLoader)

Evaluate the given context vector using the fitness function.

Parameters:

context_vector: np.ndarray

Solution of the complete problem.

data: DataLoader

Container with process data and training and test sets.

Returns:

fitness: float

Quality of the context vector.

class pyccea.fitness.SubsetSizePenalty(evaluator: WrapperEvaluation, weights: list)

Bases: WrapperFitnessFunction

Objective function that penalizes large subsets of features.

Rashid, A.N.M Bazlur, et al. “A novel penalty-based wrapper objective function for feature selection in Big Data using cooperative co-evolution.” IEEE Access 8 (2020): 150113-150129.

Attributes:

w1: float

Predictive performance weight.

w2: float

Penalty weight.

Methods

evaluate(context_vector, data)

Evaluate the given context vector using the fitness function.

evaluate(context_vector: ndarray, data: DataLoader)

Evaluate the given context vector using the fitness function.

Parameters:

context_vector: np.ndarray

Solution of the complete problem.

data: DataLoader

Container with process data and training and test sets.

Returns:

fitness: float

Quality of the context vector.

class pyccea.fitness.WrapperFitnessFunction(evaluator: WrapperEvaluation)

Bases: ABC

An abstract class for a wrapper objective function.

It measures the quality of a solution according to the predictive performance of a machine learning model.

Attributes:

evaluator: object of one of the evaluation classes

Responsible for evaluating individuals, that is, subsets of features.

Submodules

pyccea.fitness.distance module

class pyccea.fitness.distance.DistanceBasedFitness(evaluator: WrapperEvaluation, weights: list)

Bases: WrapperFitnessFunction

Objective function that maximizes balanced accuracy based on a k-nearest neighbors classifier.

The fitness function is designed as a three-objective optimization, aimed at achieving a balance between maximizing balanced accuracy while simultaneously minimizing the average distance between instances sharing the same label and maximizing the average distance between instances with different labels.

Firouznia, Marjan, Pietro Ruiu, and Giuseppe A. Trunfio. “Adaptive cooperative coevolutionary differential evolution for parallel feature selection in high-dimensional datasets.” The Journal of Supercomputing (2023): 1-30.

Attributes:

w1: float

Predictive performance weight.

w2: float

Weight of the complement of the average distance between instances and their neighbors with the same class.

w3: float

Weight of the average distance between instances and their neighbors of different classes.

Methods

evaluate(context_vector, data)

Evaluate the given context vector using the fitness function.

evaluate(context_vector: ndarray, data: DataLoader)

Evaluate the given context vector using the fitness function.

Parameters:

context_vector: np.ndarray

Solution of the complete problem.

data: DataLoader

Container with process data and training and test sets.

Returns:

fitness: float

Quality of the context vector.

pyccea.fitness.function module

class pyccea.fitness.function.WrapperFitnessFunction(evaluator: WrapperEvaluation)

Bases: ABC

An abstract class for a wrapper objective function.

It measures the quality of a solution according to the predictive performance of a machine learning model.

Attributes:

evaluator: object of one of the evaluation classes

Responsible for evaluating individuals, that is, subsets of features.

pyccea.fitness.penalty module

class pyccea.fitness.penalty.SubsetSizePenalty(evaluator: WrapperEvaluation, weights: list)

Bases: WrapperFitnessFunction

Objective function that penalizes large subsets of features.

Rashid, A.N.M Bazlur, et al. “A novel penalty-based wrapper objective function for feature selection in Big Data using cooperative co-evolution.” IEEE Access 8 (2020): 150113-150129.

Attributes:

w1: float

Predictive performance weight.

w2: float

Penalty weight.

Methods

evaluate(context_vector, data)

Evaluate the given context vector using the fitness function.

evaluate(context_vector: ndarray, data: DataLoader)

Evaluate the given context vector using the fitness function.

Parameters:

context_vector: np.ndarray

Solution of the complete problem.

data: DataLoader

Container with process data and training and test sets.

Returns:

fitness: float

Quality of the context vector.