Feature selection wrapper

Feature selection wrapper

feature_selection(dataset, method, class_attr = NULL, exclude = NULL, ...)

Arguments

dataset	we want to do feature selection on
method	selected method of feature selection
class_attr	character. Indicates the class attribute or attributes from dataset. Must exist in it.
exclude	character. Vector of attributes to exclude from the feature selection process
...	Further arguments for method

Value

The treated dataset (either with noisy instances replaced or erased)

Examples

library("smartdata")
library("rpart")
data(ecoli1, package = "imbalance")
data(HouseVotes84, package = "mlbench")

# Extracted from FSelector::best.first.search documentation
evaluator <- function(subset) {
  k <- 5
  splits <- runif(nrow(iris))
  results = sapply(1:k, function(i) {
    test.idx <- (splits >= (i - 1) / k) & (splits < i / k)
    train.idx <- !test.idx
    test <- iris[test.idx, , drop=FALSE]
    train <- iris[train.idx, , drop=FALSE]
    tree <- rpart(FSelector::as.simple.formula(subset, "Species"), train)
    error.rate = sum(test$Species != predict(tree, test, type="c")) / nrow(test)
    return(1 - error.rate)
  })
  print(subset)
  print(mean(results))
  return(mean(results))
}



super_iris <- feature_selection(iris, "Boruta", class_attr = "Species")
super_iris <- feature_selection(iris, "chi_squared",
                                class_attr = "Species", num_features = 3)
# Pick 3 attributes from the continuous ones
super_ecoli <- feature_selection(ecoli1, "information_gain",
                                 class_attr = "Class", num_features = 3)
super_ecoli <- feature_selection(ecoli1, "gain_ratio",
                                 class_attr = "Class", num_features = 3)
super_ecoli <- feature_selection(ecoli1, "sym_uncertainty",
                                 class_attr = "Class", num_features = 3)
super_votes <- feature_selection(HouseVotes84, "oneR", exclude = c("V1", "V2"),
                                 class_attr = "Class", num_features = 3)
super_votes <- feature_selection(iris, "RF_importance", class_attr = "Species",
                                 num_features = 3, type = 2)
# \donttest{
super_iris  <- feature_selection(iris, "best_first_search", exclude = "Species",
                                 eval_fun = evaluator)
#> [1] "Sepal.Length"
#> [1] 0.674
#> [1] "Sepal.Width"
#> [1] 0.491
#> [1] "Petal.Length"
#> [1] 0.944
#> [1] "Petal.Width"
#> [1] 0.959
#> [1] "Sepal.Length" "Petal.Width" 
#> [1] 0.952
#> [1] "Sepal.Width" "Petal.Width"
#> [1] 0.951
#> [1] "Petal.Length" "Petal.Width" 
#> [1] 0.942
super_iris  <- feature_selection(iris, "forward_search", exclude = "Species",
                                 eval_fun = evaluator)
#> [1] "Sepal.Length"
#> [1] 0.694
#> [1] "Sepal.Width"
#> [1] 0.505
#> [1] "Petal.Length"
#> [1] 0.927
#> [1] "Petal.Width"
#> [1] 0.954
#> [1] "Sepal.Length" "Petal.Width" 
#> [1] 0.957
#> [1] "Sepal.Width" "Petal.Width"
#> [1] 0.951
#> [1] "Petal.Length" "Petal.Width" 
#> [1] 0.936
#> [1] "Sepal.Length" "Sepal.Width"  "Petal.Width" 
#> [1] 0.951
#> [1] "Sepal.Length" "Petal.Length" "Petal.Width" 
#> [1] 0.934
super_iris  <- feature_selection(iris, "backward_search", exclude = "Species",
                                 eval_fun = evaluator)
#> [1] "Sepal.Length" "Sepal.Width"  "Petal.Length" "Petal.Width" 
#> [1] 0.939
#> [1] "Sepal.Width"  "Petal.Length" "Petal.Width" 
#> [1] 0.937
#> [1] "Sepal.Length" "Petal.Length" "Petal.Width" 
#> [1] 0.926
#> [1] "Sepal.Length" "Sepal.Width"  "Petal.Width" 
#> [1] 0.96
#> [1] "Sepal.Length" "Sepal.Width"  "Petal.Length"
#> [1] 0.944
#> [1] "Sepal.Width" "Petal.Width"
#> [1] 0.958
#> [1] "Sepal.Length" "Petal.Width" 
#> [1] 0.951
#> [1] "Sepal.Length" "Sepal.Width" 
#> [1] 0.721
# }
super_iris  <- feature_selection(iris, "cfs", class_attr = "Species")
super_iris  <- feature_selection(iris, "consistency", class_attr = "Species")