Fix #88 (workaround unexpected behavior when non-presence indicated b…

…y NA)

NEWS.md

@@ -1,10 +1,11 @@
 # Version 1.0.1
 
-*In development*
+2021-01-04
 
 Bug fixes:
 - Filtering by degree when using non-default mode and `intersections='all'` now correctly accounts for all observations (#89)
 - Empty sets/groups are now correctly removed when filtering with a non-default mode (#90)
+- Missing values are now converted to FALSE and a warning is issued to the user rather than causing an undefined behavior (#88)
 
 # Version 1.0.0
 

R/data.R

@@ -311,6 +311,11 @@ upset_data = function(
         data[, non_logical] = sapply(data[, non_logical], as.logical)
     }
 
+    if (any(is.na(data[, intersect]))) {
+        warning('Detected missing values in the columns indicating sets, coercing to FALSE')
+        data[, intersect][is.na(data[, intersect])] = FALSE
+    }
+
     intersect_in_order_of_data = colnames(data)[colnames(data) %in% intersect]
 
     non_sanitized_labels = intersect

tests/testthat/test-other-visual.R

@@ -258,4 +258,26 @@ test_that("Empty sets are removed during filtering with non-default mode", {
             )
         )
     )
+})
+
+
+test_that("Missing values are converted to FALSE", {
+    # see https://github.com/krassowski/complex-upset/issues/88
+    df = data.frame(
+        set_a = c(1, 1, 1, NA),
+        set_b = c(1, NA, NA, NA),
+        set_c = c(NA, NA, 4, NA),
+        set_d = c(NA, NA, NA, 1),
+        category = c("E", "F", "F", "G")
+    )
+
+    expect_warning(
+        upset(df, colnames(df)[1:2], min_degree=1),
+        regexp='Detected missing values in the columns indicating sets, coercing to FALSE'
+    )
+
+    expect_doppelganger(
+        "Degrees are filtered (min_degree=1) even if user indicated non-presence by NA",
+        upset(df, colnames(df)[1:2], min_degree=1)
+    )
 })