fixed a bug in using BoxCox

For issue #177

.gitignore

@@ -53,3 +53,7 @@ release_process/Open_Data_Science_Conference/caret.snm
 release_process/Open_Data_Science_Conference/caret.toc
 
 release_process/Open_Data_Science_Conference/caret.vrb
+
+models/files/.RData
+
+models/parseModels.Rout

pkg/caret/R/preProcess.R

@@ -70,7 +70,7 @@ preProcess.default <- function(x, method = c("center", "scale"),
   if(any(method == "BoxCox"))
   {
     if(verbose) cat("Estimating Box-Cox transformations for the predictors...")
-    bc <- lapply(x,
+    bc <- lapply(as.data.frame(x),
                  BoxCoxTrans,
                  fudge = fudge,
                  na.rm = na.remove,
@@ -97,7 +97,7 @@ preProcess.default <- function(x, method = c("center", "scale"),
       out
     }
     if(verbose) cat("Estimating Yeo-Johnson transformations for the predictors...")
-    yj <- lapply(x, yjWrap, numUnique = numUnique)
+    yj <- lapply(as.data.frame(x), yjWrap, numUnique = numUnique)
     if(verbose) cat(" applying them to training data\n")
     ## Find a better way of doing this
     lam <- unlist(lapply(yj, function(x) if(class(x) == "powerTransform") x$lambda else NA))
@@ -115,7 +115,7 @@ preProcess.default <- function(x, method = c("center", "scale"),
   if(any(method == "expoTrans"))
   {
     if(verbose) cat("Estimating exponential transformations for the predictors...")
-    et <- lapply(x, expoTrans.default, numUnique = numUnique)
+    et <- lapply(as.data.frame(x), expoTrans.default, numUnique = numUnique)
     if(verbose) cat(" applying them to training data\n")
     for(i in seq(et)) x[,i] <- predict(et[[i]], x[,i])
   } else et <- NULL  

pkg/caret/tests/testthat/test_BoxCox.R

@@ -0,0 +1,31 @@
+context('Box Cox transformations')
+
+###################################################################
+## Generate data and do BC using the source function to get 
+## expected results
+
+library(MASS)
+
+set.seed(1)
+dat <- matrix(runif(30), ncol = 3)
+dat[,1] <- exp(dat[,1])
+colnames(dat) <- paste0("x", 1:3)
+
+exp_lambdas <- rep(NA, 3)
+for(i in 1:ncol(dat)) {
+  tmp <- as.data.frame(dat)[,i,drop = FALSE]
+  names(tmp)[1] <- "x"
+  tmp_bc <- boxcox(x ~ 1, data = tmp, plotit = FALSE, lambda = seq(-2, 2, by = .1))
+  exp_lambdas[i] <- tmp_bc$x[which.max(tmp_bc$y)]
+}
+
+check_BoxCox <- function(x, expected = NULL) {
+  pp1 <- preProcess(x, method = "BoxCox")
+  obs_lambdas1 <- unlist(lapply(pp1$bc, function(x) x$lambda))
+  names(obs_lambdas1) <- NULL
+  expect_equal(obs_lambdas1, expected)
+}
+
+check_BoxCox(dat, expected = exp_lambdas)
+check_BoxCox(as.data.frame(dat), expected = exp_lambdas)
+