R/R6_Model.R

   1 #' @title R6 class representing a (generic) model.
   2 #'
   3 #' @description
   4 #' "Model" class, containing a (generic) learning function, which from
   5 #' data + target [+ params] returns a prediction function X --> y.
   6 #' Parameters for cross-validation are either provided or estimated.
   7 #' Model family can be chosen among "rf", "tree", "ppr" and "knn" for now.
   8 #'
   9 #' @importFrom FNN knn.reg
  10 #' @importFrom class knn
  11 #' @importFrom stats ppr
  12 #' @importFrom randomForest randomForest
  13 #' @importFrom rpart rpart
  14 #' @importFrom caret var_seq
  15 #'
  16 #' @export
  17 Model <- R6::R6Class("Model",
  18   public = list(
  19     #' @field nmodels Number of parameters (= number of [predictive] models)
  20     nmodels = NA,
  21     #' @description Create a new generic model.
  22     #' @param data Matrix or data.frame
  23     #' @param target Vector of targets (generally numeric or factor)
  24     #' @param task "regression" or "classification"
  25     #' @param gmodel Generic model returning a predictive function; chosen
  26     #'               automatically given data and target nature if not provided.
  27     #' @param params List of parameters for cross-validation (each defining a model)
  28     initialize = function(data, target, task, gmodel = NULL, params = NULL) {
  29       if (is.null(gmodel)) {
  30         # (Generic) model not provided
  31         all_numeric <- is.numeric(as.matrix(data))
  32         if (!all_numeric)
  33           # At least one non-numeric column: use random forests or trees
  34           # TODO: 4 = arbitrary magic number...
  35           gmodel = ifelse(ncol(data) >= 4, "rf", "tree")
  36         else
  37           # Numerical data
  38           gmodel = ifelse(task == "regression", "ppr", "knn")
  39       }
  40       if (is.null(params))
  41         # Here, gmodel is a string (= its family),
  42         # because a custom model must be given with its parameters.
  43         params <- as.list(private$getParams(gmodel, data, target))
  44       private$params <- params
  45       if (is.character(gmodel))
  46         gmodel <- private$getGmodel(gmodel, task)
  47       private$gmodel <- gmodel
  48       self$nmodels <- length(private$params)
  49     },
  50     #' @description
  51     #' Returns the model at index "index", trained on dataHO/targetHO.
  52     #' index is between 1 and self$nmodels.
  53     #' @param dataHO Matrix or data.frame
  54     #' @param targetHO Vector of targets (generally numeric or factor)
  55     #' @param index Index of the model in 1...nmodels
  56     get = function(dataHO, targetHO, index) {
  57       private$gmodel(dataHO, targetHO, private$params[[index]])
  58     }
  59   ),
  60   private = list(
  61     # No need to expose model or parameters list
  62     gmodel = NULL,
  63     params = NULL,
  64     # Main function: given a family, return a generic model, which in turn
  65     # will output a predictive model from data + target + params.
  66     getGmodel = function(family, task) {
  67       if (family == "tree") {
  68         function(dataHO, targetHO, param) {
  69           require(rpart)
  70           method <- ifelse(task == "classification", "class", "anova")
  71           df <- data.frame(cbind(dataHO, target=targetHO))
  72           model <- rpart::rpart(target ~ ., df, method=method, control=list(cp=param))
  73           function(X) predict(model, X)
  74         }
  75       }
  76       else if (family == "rf") {
  77         function(dataHO, targetHO, param) {
  78           require(randomForest)
  79           if (task == "classification" && !is.factor(targetHO))
  80             targetHO <- as.factor(targetHO)
  81           model <- randomForest::randomForest(dataHO, targetHO, mtry=param)
  82           function(X) predict(model, X)
  83         }
  84       }
  85       else if (family == "ppr") {
  86         function(dataHO, targetHO, param) {
  87           model <- stats::ppr(dataHO, targetHO, nterms=param)
  88           function(X) predict(model, X)
  89         }
  90       }
  91       else if (family == "knn") {
  92         if (task == "classification") {
  93           function(dataHO, targetHO, param) {
  94             require(class)
  95             function(X) class::knn(dataHO, X, cl=targetHO, k=param)
  96           }
  97         }
  98         else {
  99           function(dataHO, targetHO, param) {
 100             require(FNN)
 101             function(X) FNN::knn.reg(dataHO, X, y=targetHO, k=param)$pred
 102           }
 103         }
 104       }
 105     },
 106     # Return a default list of parameters, given a gmodel family
 107     getParams = function(family, data, target) {
 108       if (family == "tree") {
 109         # Run rpart once to obtain a CV grid for parameter cp
 110         require(rpart)
 111         df <- data.frame(cbind(data, target=target))
 112         ctrl <- list(
 113           minsplit = 2,
 114           minbucket = 1,
 115           maxcompete = 0,
 116           maxsurrogate = 0,
 117           usesurrogate = 0,
 118           xval = 0,
 119           surrogatestyle = 0,
 120           maxdepth = 30)
 121         r <- rpart(target ~ ., df, method="class", control=ctrl)
 122         cps <- r$cptable[-1,1]
 123         if (length(cps) <= 11)
 124           return (cps)
 125         step <- (length(cps) - 1) / 10
 126         cps[unique(round(seq(1, length(cps), step)))]
 127       }
 128       else if (family == "rf") {
 129         p <- ncol(data)
 130         # Use caret package to obtain the CV grid of mtry values
 131         require(caret)
 132         caret::var_seq(p, classification = (task == "classificaton"),
 133                        len = min(10, p-1))
 134       }
 135       else if (family == "ppr")
 136         # This is nterms in ppr() function
 137         1:10
 138       else if (family == "knn") {
 139         n <- nrow(data)
 140         # Choose ~10 NN values
 141         K <- length(unique(target))
 142         if (n <= 10)
 143           return (1:(n-1))
 144         sqrt_n <- sqrt(n)
 145         step <- (2*sqrt_n - 1) / 10
 146         grid <- unique(round(seq(1, 2*sqrt_n, step)))
 147         if (K == 2) {
 148           # Common binary classification case: odd number of neighbors
 149           for (i in 2:11) {
 150             if (grid[i] %% 2 == 0)
 151               grid[i] <- grid[i] + 1 #arbitrary choice
 152           }
 153         }
 154         grid
 155       }
 156     }
 157   )
 158 )