|
| 1 | +#' @title Classification Boosting Learner |
| 2 | +#' @author annanzrv |
| 3 | +#' @name mlr_learners_classif.adabag |
| 4 | +#' |
| 5 | +#' @description |
| 6 | +#' Classification boosting algorithm. |
| 7 | +#' Calls [adabag::boosting()] from \CRANpkg{adabag}. |
| 8 | +#' |
| 9 | +#' @section Initial parameter values: |
| 10 | +#' - `xval`: |
| 11 | +#' * Actual default: 10L |
| 12 | +#' * Initial value: 0L |
| 13 | +#' * Reason for change: Set to 0 for speed. |
| 14 | +#' |
| 15 | +#' @references |
| 16 | +#' `r format_bib("adabag2013")` |
| 17 | +#' |
| 18 | +#' @templateVar id classif.adabag |
| 19 | +#' @template learner |
| 20 | +#' |
| 21 | +#' |
| 22 | +#' @template seealso_learner |
| 23 | +#' @template example |
| 24 | +#' @export |
| 25 | +LearnerClassifAdabag = R6Class("LearnerClassifAdabag", |
| 26 | + inherit = LearnerClassif, |
| 27 | + public = list( |
| 28 | + #' @description |
| 29 | + #' Creates a new instance of this [R6][R6::R6Class] class. |
| 30 | + initialize = function() { |
| 31 | + param_set = ps( |
| 32 | + boos = p_lgl(default = TRUE, tags = "train"), |
| 33 | + coeflearn = p_fct(default = "Breiman", levels = c("Breiman", "Freund", "Zhu"), tags = "train"), |
| 34 | + cp = p_dbl(default = 0.01, lower = 0, upper = 1, tags = "train"), |
| 35 | + maxcompete = p_int(default = 4L, lower = 0L, tags = "train"), |
| 36 | + maxdepth = p_int(default = 30L, lower = 1L, upper = 30L, tags = "train"), |
| 37 | + maxsurrogate = p_int(default = 5L, lower = 0L, tags = "train"), |
| 38 | + mfinal = p_int(default = 100L, lower = 1L, tags = "train"), |
| 39 | + minbucket = p_int(lower = 1L, tags = "train"), |
| 40 | + minsplit = p_int(default = 20L, lower = 1L, tags = "train"), |
| 41 | + newmfinal = p_int(tags = "predict"), |
| 42 | + surrogatestyle = p_int(default = 0L, lower = 0L, upper = 1L, tags = "train"), |
| 43 | + usesurrogate = p_int(default = 2L, lower = 0L, upper = 2L, tags = "train"), |
| 44 | + xval = p_int(default = 0L, lower = 0L, tags = "train") |
| 45 | + ) |
| 46 | + param_set$values = list(xval = 0L) |
| 47 | + |
| 48 | + super$initialize( |
| 49 | + id = "classif.adabag", |
| 50 | + packages = c("adabag", "rpart"), |
| 51 | + feature_types = c("integer", "numeric", "factor"), |
| 52 | + predict_types = c("response", "prob"), |
| 53 | + param_set = param_set, |
| 54 | + properties = c("importance", "missings", "multiclass", "twoclass"), |
| 55 | + man = "mlr3extralearners::mlr_learners_classif.adabag", |
| 56 | + label = "Adabag Boosting" |
| 57 | + ) |
| 58 | + }, |
| 59 | + #' @description |
| 60 | + #' The importance scores are extracted from the model. |
| 61 | + #' @return Named `numeric()`. |
| 62 | + importance = function() { |
| 63 | + if (is.null(self$model)) { |
| 64 | + stopf("No model stored") |
| 65 | + } |
| 66 | + sort(self$model$importance, decreasing = TRUE) |
| 67 | + } |
| 68 | + ), |
| 69 | + |
| 70 | + private = list( |
| 71 | + .train = function(task) { |
| 72 | + # get parameters for training |
| 73 | + pars = self$param_set$get_values(tags = "train") |
| 74 | + |
| 75 | + args_ctrl = formalArgs(rpart::rpart.control) |
| 76 | + pars_ctrl = pars[names(pars) %in% args_ctrl] |
| 77 | + |
| 78 | + # Create rpart control object |
| 79 | + ctrl = invoke( |
| 80 | + rpart::rpart.control, |
| 81 | + .args = pars_ctrl |
| 82 | + ) |
| 83 | + |
| 84 | + # Remove rpart control parameters from pars |
| 85 | + pars = pars[names(pars) %nin% args_ctrl] |
| 86 | + |
| 87 | + # Add control to pars |
| 88 | + pars$control = ctrl |
| 89 | + |
| 90 | + # Get formula and data |
| 91 | + formula = task$formula() |
| 92 | + data = task$data() |
| 93 | + |
| 94 | + # Train model |
| 95 | + invoke(adabag::boosting, |
| 96 | + formula = formula, |
| 97 | + data = data, |
| 98 | + .args = pars |
| 99 | + ) |
| 100 | + }, |
| 101 | + .predict = function(task) { |
| 102 | + # get parameters with tag "predict" |
| 103 | + pars = self$param_set$get_values(tags = "predict") |
| 104 | + |
| 105 | + # get newdata and ensure same ordering in train and predict |
| 106 | + newdata = ordered_features(task, self) |
| 107 | + |
| 108 | + # Calculate predictions for the selected predict type |
| 109 | + type = self$predict_type |
| 110 | + |
| 111 | + # adaboost requires target column |
| 112 | + newdata[, "target"] = factor(rep(1, nrow(newdata)), levels = task$class_names) |
| 113 | + |
| 114 | + pred = invoke(predict, self$model, newdata = newdata, .args = pars) |
| 115 | + |
| 116 | + if (type == "prob") { |
| 117 | + # Ensure probabilities are ordered according to task class levels |
| 118 | + prob = mlr3misc::set_col_names(pred$prob, task$class_names) |
| 119 | + list(prob = prob) |
| 120 | + } else { |
| 121 | + # Create response factor with correct levels |
| 122 | + response = factor(pred$class, levels = task$class_names) |
| 123 | + list(response = response) |
| 124 | + } |
| 125 | + } |
| 126 | + ) |
| 127 | +) |
| 128 | + |
| 129 | +.extralrns_dict$add("classif.adabag", LearnerClassifAdabag) |
0 commit comments