mlRpart method added

Author: phgrosjean

DESCRIPTION

@@ -1,7 +1,6 @@
 Package: mlearning
 Type: Package
-Version: 1.1.1
-Date: 2022-04-26
+Version: 1.2.0
 Title: Machine Learning Algorithms with Unified Interface and Confusion Matrices
 Description: A unified interface is provided to various machine learning
   algorithms like like linear or quadratic discriminant analysis, k-nearest
@@ -20,7 +19,7 @@ Authors@R: c(
     email = "kevin.denis@umons.ac.be"))
 Maintainer: Philippe Grosjean <phgrosjean@sciviews.org>
 Depends: R (>= 3.0.4)
-Imports: stats, grDevices, class, nnet, MASS, e1071, randomForest, ipred
+Imports: stats, grDevices, class, nnet, MASS, e1071, randomForest, ipred, rpart
 Suggests: mlbench, datasets, RColorBrewer, spelling, knitr, rmarkdown, covr
 URL: https://www.sciviews.org/mlearning/
 BugReports: https://github.com/SciViews/mlearning/issues

NAMESPACE

@@ -9,6 +9,7 @@ import(MASS)
 import(e1071)
 import(randomForest)
 import(ipred)
+import(rpart)
 #import(klaR)
 #import(RWeka)
 
@@ -19,6 +20,7 @@ export(mlRforest)
 export(mlKnn)
 export(mlLvq)
 export(mlNnet)
+export(mlRpart)
 export(mlSvm)
 export(mlNaiveBayes)
 #export(mlNaiveBayesWeka)
@@ -67,6 +69,7 @@ S3method(mlRforest, default)
 S3method(mlKnn, default)
 S3method(mlLvq, default)
 S3method(mlNnet, default)
+S3method(mlRpart, default)
 S3method(mlSvm, default)
 S3method(mlNaiveBayes, default)
 #S3method(mlNaiveBayesWeka, default)
@@ -78,6 +81,7 @@ S3method(mlKnn, formula)
 S3method(mlLvq, formula)
 S3method(mlNnet, formula)
 S3method(mlSvm, formula)
+S3method(mlRpart, formula)
 S3method(mlNaiveBayes, formula)
 #S3method(mlNaiveBayesWeka, formula)
 
@@ -88,4 +92,5 @@ S3method(predict, mlKnn)
 S3method(predict, mlLvq)
 S3method(predict, mlNnet)
 S3method(predict, mlSvm)
+S3method(predict, mlRpart)
 S3method(predict, mlNaiveBayes)

NEWS.md

@@ -1,3 +1,7 @@
+# mlearning 1.2.0
+
+-   The `mlRpart()` function implements `rpart::rpart()` for using decision trees.
+
 # mlearning 1.1.1
 
 -   The description is extended.

R/mlearning.R

@@ -626,6 +626,68 @@ prior = object$prior, method = c("plug-in", "predictive", "debiased", "looCV",
     stop("unrecognized 'type' (must be 'class', 'membership' or 'both')"))
 }
 
+# rpart() from {rpart}
+mlRpart <- function(train, ...)
+  UseMethod("mlRpart")
+
+mlRpart.formula <- function(formula, data, ..., subset, na.action) {
+  mlearning(formula, data = data, method = "mlRpart", model.args =
+    list(formula  = formula, data = substitute(data),
+        subset = substitute(subset)), call = match.call(), ...,
+    subset = subset, na.action = substitute(na.action))
+}
+
+mlRpart.default <- function(train, response, ..., .args. = NULL) {
+  dots <- list(...)
+  if (is.null(.args.) || !length(.args.)) {
+    if (!length(response)) {# unsupervised
+      stop("Unsupervised classification is not possible with mlRpart(), see hclust() instead.")
+    } else if (is.factor(response)) {
+      type <- "classification"
+    } else type <- "regression"
+
+    .args. <- list(levels = levels(response),
+      n = c(intial = NROW(train), final = NROW(train)),
+      type = type, na.action = "na.pass",
+      mlearning.call = match.call(), method = "mlRpart")
+  }
+
+  # Combine train + response and use the formula interface which is the only one
+  data <- train
+  data$.class <- response
+  res <- rpart::rpart(.class ~ ., data = data, ...)
+  res$predicted <- predict(res, type = "class")
+
+  # Return a mlearning object
+  structure(res, formula = .args.$formula, train = train,
+    response = response, levels = .args.$levels, n = .args.$n, args = dots,
+    optim = .args.$optim, numeric.only = FALSE, type = .args.$type,
+    pred.type = c(class = "class", membership = "prob"),
+    summary = NULL, na.action = .args.$na.action,
+    mlearning.call = .args.$mlearning.call, method = .args.$method,
+    algorithm = "recursive partitioning tree",
+    class = c("mlRpart", "mlearning", "rpart"))
+}
+
+predict.mlRpart <- function(object, newdata,
+type = c("class", "membership", "both"),
+  method = c("direct", "cv"), ...) {
+
+  type <- as.character(type)[1]
+
+  # If method == "cv", delegate to cvpredict()
+  method <- as.character(method)[1]
+  if (method == "cv") {
+    if (!missing(newdata))
+      stop("cannot handle new data with method = 'cv'")
+    return(cvpredict(object = object, type = type, ...))
+  } else {
+    predict.mlearning(object = object, newdata = newdata,
+      type = type, ...)
+  }
+}
+
+
 mlRforest <- function(train, ...)
   UseMethod("mlRforest")
 

man/mlearning.Rd

@@ -17,6 +17,10 @@
 \alias{mlQda.default}
 \alias{mlQda.formula}
 \alias{predict.mlQda}
+\alias{mlRpart}
+\alias{mlRpart.default}
+\alias{mlRpart.formula}
+\alias{predict.mlRpart}
 \alias{mlRforest}
 \alias{mlRforest.default}
 \alias{mlRforest.formula}
@@ -81,6 +85,12 @@ mlQda(train, ...)
     prior = object$prior, method = c("plug-in", "predictive", "debiased",
     "looCV", "cv"), ...)
 
+mlRpart(train, ...)
+\method{mlRpart}{default}(train, response, ..., .args. = NULL)
+\method{mlRpart}{formula}(formula, data, ..., subset, na.action)
+\method{predict}{mlRpart}(object, newdata, type = c("class", "membership", "both"),
+  method = c("direct", "cv"), ...)
+
 mlRforest(train, ...)
 \method{mlRforest}{default}(train, response, ntree = 500, mtry, replace = TRUE, classwt = NULL, ...)
 \method{mlRforest}{formula}(formula, data, ntree = 500, mtry, replace = TRUE, classwt = NULL, ...,
@@ -217,6 +227,7 @@ train(object, ...)
   \item{eps}{ double for specifying an epsilon-range to apply laplace smoothing
     (to replace zero or close-zero probabilities by 'threshold') for Naive Bayes
     predictions. }
+  \item{.args.}{ Used internally, do not provide anything here. }
 }
 
 \value{
@@ -308,6 +319,21 @@ summary(res <- mlQda(Class ~ ., data = HouseVotes84, na.action = na.omit))
 confusion(res) # Self-consistency
 
 
+## Supervised classification using rpart
+summary(res <- mlRpart(Species ~ ., data = irisTrain))
+plot(res, margin = 0.03, uniform = TRUE)
+text(res, use.n = FALSE)
+predict(res) # Default type is class
+predict(res, type = "membership")
+predict(res, type = "both")
+confusion(res) # Self-consistency
+## Cross-validation prediction is a good choice when there is no test set:
+predict(res, method = "cv")  # Idem: cvpredict(res)
+confusion(res, method = "cv") # Cross-validation for performances estimation
+## Evaluation of performances using a separate test set
+confusion(predict(res, newdata = irisTest), irisTest$Species) # Test set perfs
+
+
 ## Supervised classification using random forest
 summary(res <- mlRforest(Species ~ ., data = irisTrain))
 plot(res)

man/mlearning.package.Rd

@@ -13,8 +13,8 @@
   \tabular{ll}{
     Package: \tab mlearning\cr
     Type: \tab Package\cr
-    Version: \tab 1.0.3\cr
-    Date: \tab 2017-12-12\cr
+    Version: \tab 1.2.0\cr
+    Date: \tab 2022-09-08\cr
     License: \tab GPL 2 or above at your convenience.\cr
   }
 }