evolve: Evolve a Population of Models Over Generations
In predomics/predomicspkg: Interpretable Prediction in Omics Data

evolve

R Documentation

Evolve a Population of Models Over Generations

Description

This function evolves a population of models across a specified number of generations, using selection, crossover, and mutation operations. It tracks the best-performing individuals in each generation to encourage convergence toward optimal solutions.

Usage

evolve(X, y, clf, pop, seed = NULL)

Arguments

`X`	A matrix or data frame representing the features of the training dataset.
`y`	A vector representing the target variable for the training dataset.
`clf`	A classifier object containing parameters for the evolutionary algorithm, including: - 'nb_generations': The number of generations to evolve the population. - 'size_pop': The target population size. - 'current_sparsity': The desired sparsity level for individuals. - 'select_type': Selection method, which can be "elite," "tournoi," or "mixed." - 'select_perc1' and 'select_perc2': Percentages used in mixed selection. - 'mutate_size': Percentage of the population to mutate in each generation. - 'plot': Boolean indicating whether to plot the evolution of the best score. - 'convergence': Boolean indicating if convergence testing should be applied. - 'convergence_steps': The number of steps used for convergence testing.
`pop`	A list representing the initial population of models, with each individual represented as a vector of selected features.
`seed`	Optional integer seed for random number generation to ensure reproducibility.

Details

The function evolves a population of models over multiple generations using the following steps: 1. **Evaluation**: Evaluates each individual in the population. 2. **Selection**: Selects parents based on specified selection criteria ('select_type'). 3. **Crossover**: Combines genetic material from selected parents to produce children. 4. **Mutation**: Mutates a portion of the population to introduce genetic diversity. 5. **Convergence Testing**: Optionally tests for convergence, stopping the evolution early if stability is detected in performance. **Population Structure**: The population is a list of vectors, where each vector represents a sparse model (a subset of features). After each generation, the function retains the best-performing individuals, tracking the evolution of the best score across generations.

Value

A list representing the final evolved population of models.

Examples

## Not run: 
clf <- list(
  params = list(nb_generations = 10, size_pop = 20, current_sparsity = 5, 
                select_type = "elite", mutate_size = 10, plot = TRUE, convergence = FALSE)
)
X <- matrix(runif(200), nrow = 20)
y <- sample(0:1, 20, replace = TRUE)
initial_pop <- list(1:5, 2:6, 3:7)  # Example initial population
final_pop <- evolve(X, y, clf, initial_pop, seed = 42)
print(final_pop)

## End(Not run)

predomics/predomicspkg documentation built on Dec. 11, 2024, 11:06 a.m.