DEbest

DE/best/k variant for discriminative Bayesian Network parameter learning

DEbest implements the DE/best/1 and DE/best/2 mutation strategies with either binomial or exponential crossover.

Usage

DEbest(
  NP = 40,
  G = 100,
  data,
  class.name,
  F = 0.5,
  CR = 0.7,
  mutation.pairs = c(1, 2),
  crossover = c("bin", "exp"),
  structure = c("nb", "tancl", "tan", "hc"),
  edgelist = NULL,
  verbose = 25,
  ...
)

Parameters

NP · integer · default: "40"

Population size. Must be > 5; values ≤ 5 are reset to 40.

G · integer · default: "100"

Maximum number of generations. Values ≤ 1 are reset to 100.

data · data.frame · required

Data frame from which to learn the classifier. Must contain the class variable.

class.name · character · required

Name of the class variable column in data.

F · numeric · default: "0.5"

Mutation factor. Must be in [0, 2]; values outside this range are reset to 0.5.

CR · numeric · default: "0.7"

Crossover rate. Must be in [0, 1]; values outside this range are reset to 0.7.

mutation.pairs · integer · default: "1"

Number of difference vector pairs in mutation: 1 (DE/best/1) or 2 (DE/best/2).

crossover · character · default: "bin"

Crossover type: "bin" (binomial) or "exp" (exponential).

structure · character · default: "nb"

BN structure learning method: "nb" (Naive Bayes), "tan" or "tancl" (TAN via Chow-Liu), "hc" (Hill-Climbing).

edgelist · matrix

Optional custom BN topology as an edges × 2 matrix with columns (from, to). Overrides structure.

verbose · integer · default: "25"

Print progress every this many generations. Set to 0 to suppress output.

...

Additional arguments passed to bnclassify::tan_cl or bnclassify::tan_hc (e.g., k for hill-climbing folds).

Return value

An object of class DE containing:

Best · bnc_bn

The Bayesian Network with the highest CLL in the final population.

BestCLL · numeric

Conditional Log-Likelihood of the best network.

pobFinal · list

All bnc_bn objects in the final population.

CLLPobFinal · numeric vector

CLL values for each member of the final population.

N.evals · integer

Total number of CLL evaluations performed.

convergence · numeric vector

Best CLL recorded at each generation.

evaluations · integer vector

Cumulative evaluations at each generation.

Examples

library(dplbnDE)
data(car)

# DE/best/1/bin with TAN structure
result <- DEbest(
  NP = 20, G = 25,
  data = car,
  class.name = names(car)[7],
  F = 0.5, CR = 0.7,
  mutation.pairs = 1,
  crossover = "bin",
  structure = "tan",
  verbose = 5
)

print(result)
plot(result)

# Predict using the best network
predictions <- predict(result$Best, car)
accuracy(predictions, car$class)