Add logic to handle column of NAs

mice Imputation code would fail if a data.frame with a column of missing NA variables is passed in. Logic added to catch and remove a column of NAs and notify user of the removal with a warning

DESCRIPTION

@@ -1,19 +1,17 @@
 Package: flipImputation
 Type: Package
 Title: Data frame imputation
-Version: 1.0.0
+Version: 1.0.1
 Author: Displayr <opensource@displayr.com>
 Maintainer: Displayr <opensource@displayr.com>
 Description: Functions for imputing data using mice and hot.deck.
 License: GPL-3
 LazyData: TRUE
-Imports:
-    hot.deck,
-    mice,
-    flipU
-Suggests:
-    flipExampleData,
-    testthat
-Remotes:
-    Displayr/flipU
-RoxygenNote: 6.1.0
+Imports: hot.deck,
+ mice,
+ flipU
+Suggests: flipExampleData,
+ testthat
+Remotes: Displayr/flipU
+Encoding: UTF-8
+RoxygenNote: 7.0.0

R/imputation.R

@@ -30,6 +30,14 @@ Imputation <- function(data = NULL, formula = NULL, method = "try mice", m = 1,
         }
         FALSE
     }
+    missing.variables <- apply(data, 2, function(x) all(is.na(x)))
+    if(any(missing.variables))
+    {
+        missing.variable.names <- paste0(names(data[missing.variables]), collapse = ", ")
+        warning("Data has variable(s) that are entirely missing values (all observed values of the variable are missing). ",
+                "These variable(s) have been removed from the analysis (", missing.variable.names, ").")
+        data <- data[, !missing.variables]
+    }
     if(!any(is.na(data)))
         return(lapply(seq(m), function(x) data))
 
@@ -54,12 +62,6 @@ Imputation <- function(data = NULL, formula = NULL, method = "try mice", m = 1,
         dat.colnames <- NULL
         imputed.data <- suppressWarnings(try(
             {
-                # Require is used instead of Depends because using Depends
-                # implies that all downstream packages must also use Depends.
-                # This can substantially increase the load time
-                # as well as risk of conflicting names when the hierachy of
-                # dependencies is very deep
-                require("mice")
                 set.seed(seed)
                 dat.colnames <- colnames(data)
                 colnames(data) <- paste0("A", 1:ncol(data)) # need to replace names to avoid errors in mice v3.0.0

tests/testthat/test-imputation.R

@@ -61,6 +61,14 @@ test_that("Errors with missing values",
 {
     d <- data.frame(y = 1:10, x = 1:10)
     expect_warning(Imputation(d, y ~ x, m = 1, seed = 1233), NA) # nothing missing
+    d$z <- rep(NA, nrow(d))
+    expect_warning(Imputation(d, y ~ ., m = 1, seed = 1233),
+                   "Data has variable(s) that are entirely missing values (all observed values of the variable are missing). These variable(s) have been removed from the analysis (z).", fixed = TRUE)
+    d$w <- d$z
+    expect_warning(Imputation(d, y ~ ., m = 1, seed = 1233),
+                   "Data has variable(s) that are entirely missing values (all observed values of the variable are missing). These variable(s) have been removed from the analysis (z, w).", fixed = TRUE)
+    d$z <- NULL
+    d$w <- NULL
     d$y[2] <- NA
     expect_warning(Imputation(d, y ~ x, m = 1, seed = 1233),
                   "Imputation has been selected, but the data has no missing values in the predictors, so nothing has been imputed.")
@@ -1019,7 +1027,8 @@ test_that("Rownames preserved",
 
 test_that("No imputation needed",
 {
-    expect_warning(z <- Imputation(seq(100)), NA)
-    expect_equal(z[[1]], seq(100))
+    x <- data.frame(x = seq(100))
+    expect_warning(z <- Imputation(x), NA)
+    expect_equal(z[[1]], x)
 })