Improve error message for values out of support

src/glmfit.jl

@@ -846,16 +846,28 @@ function _initialeta!(eta::AbstractVector,
 end
 
 # Helper function to check that the values of y are in the allowed domain
-function checky(y, d::Distribution)
-    if any(x -> !insupport(d, x), y)
-        throw(ArgumentError("y must be in the support of D"))
+function checky(y::AbstractVector, d::Distribution)
+    for yy in y
+        if !insupport(d, yy)
+            minsupport, maxsupport = extrema(d)
+            left = insupport(d, minsupport) ? '[' : '('
+            right = insupport(d, maxsupport) ? ']' : ')'
+            discrete = d isa DiscreteUnivariateDistribution && eltype(d) !== Bool ?
+                "discrete over " : ""
+            throw(ArgumentError("value $yy in response is not in the support of the " *
+                                "$(nameof(typeof(d))) distribution, which is $(discrete)" *
+                                "$left$minsupport, $maxsupport$right"))
+        end
     end
     return nothing
 end
-function checky(y, d::Binomial)
-    for yy in y
-        0 ≤ yy ≤ 1 || throw(ArgumentError("$yy in y is not in [0,1]"))
-    end
+# we allow proportions but Distributions.jl takes only the number of successes
+function checky(y::AbstractVector, d::Binomial)
+     for yy in y
+        0 ≤ yy ≤ 1 ||
+            throw(ArgumentError("value $yy in response is not in the support of " *
+                                "the Binomial distribution, which is [0, 1]"))
+     end
     return nothing
 end
 

test/runtests.jl

@@ -1702,8 +1702,6 @@ end
           [coef(lm1), stderror(lm1), coef(lm1) ./ stderror(lm1)]
     @test t.cols[4] ≈ [0.5515952883836446, 3.409192065429258e-7]
     @test hcat(t.cols[5:6]...) == confint(lm1)
-    # TODO: call coeftable(gm1, ...) directly once DataFrameRegressionModel
-    # supports keyword arguments
     t = coeftable(lm1; level=0.99)
     @test hcat(t.cols[5:6]...) == confint(lm1; level=0.99)
 
@@ -2653,3 +2651,28 @@ end
         x            -0.502451    0.675377  -0.74    0.4982   -2.3776      1.3727
         ─────────────────────────────────────────────────────────────────────────"""
 end
+
+@testset "response in support" begin
+    @test GLM.checky([0.0, -0.0, 1.0, -1.0, Inf, -Inf], Normal()) === nothing
+    @test_throws(ArgumentError("value NaN in response is not in the support of " *
+                               "the Normal distribution, which is [-Inf, Inf]"),
+                                GLM.checky([NaN], Normal()))
+
+    @test GLM.checky([0.0, -0.0, 1.0], Poisson()) === nothing
+    @test_throws(ArgumentError("value 0.5 in response is not in the support of " *
+                               "the Poisson distribution, which is discrete over [0, Inf)"),
+                                GLM.checky([0.5], Poisson()))
+    @test_throws(ArgumentError("value Inf in response is not in the support of " *
+                               "the Poisson distribution, which is discrete over [0, Inf)"),
+                                GLM.checky([Inf], Poisson()))
+
+    @test GLM.checky([0.0, -0.0, 1.0, 0.5], Binomial()) === nothing
+    @test_throws(ArgumentError("value 2.0 in response is not in the support of " *
+                               "the Binomial distribution, which is [0, 1]"),
+                                GLM.checky([2.0], Binomial()))
+
+    @test GLM.checky([0.0, -0.0, 1.0], Bernoulli()) === nothing
+    @test_throws(ArgumentError("value 0.5 in response is not in the support of " *
+                               "the Bernoulli distribution, which is [false, true]"),
+                                GLM.checky([0.5], Bernoulli()))
+end