Set minimum of OrderedLogistic distribution to 1 (#2603)

* Set minimum of OrderedLogistic distribution to 1

The minimum of the OrderedLogistic function is supposed to be 1, not 0.
This is consistent with the documentation, and the behavior of the
distribution when using rand().

* Add test for support of OrderedLogistic

We add the test that the support should have the right number of
elements, and that the probabilities add up to 1.0.

* Format

* Bump patch for release

---------

Co-authored-by: Penelope Yong <penelopeysm@gmail.com>

Author: Noricc

HISTORY.md

@@ -1,3 +1,7 @@
+# Release 0.39.2
+
+Fixed a bug in the support of `OrderedLogistic` (by changing the minimum from 0 to 1).
+
 # Release 0.39.1
 
 No changes from 0.39.0 — this patch is released just to re-trigger a Documenter.jl run.

Project.toml

@@ -1,6 +1,6 @@
 name = "Turing"
 uuid = "fce5fe82-541a-59a6-adf8-730c64b5f9a0"
-version = "0.39.1"
+version = "0.39.2"
 
 [deps]
 ADTypes = "47edcb42-4c32-4615-8424-f2b9edc5f35b"

src/stdlib/distributions.jl

@@ -126,7 +126,7 @@ function OrderedLogistic(η, cutpoints::AbstractVector)
     return OrderedLogistic{typeof(η),typeof(cutpoints)}(η, cutpoints)
 end
 
-Base.minimum(d::OrderedLogistic) = 0
+Base.minimum(d::OrderedLogistic) = 1
 Base.maximum(d::OrderedLogistic) = length(d.cutpoints) + 1
 
 function Distributions.logpdf(d::OrderedLogistic, k::Real)

test/stdlib/distributions.jl

@@ -32,6 +32,17 @@ using Turing
         @test all(((x, y),) -> abs(x - y) < 0.001, zip(p, pmf))
     end
 
+    @testset "distribution functions" begin
+        d = OrderedLogistic(0, [1, 2, 3])
+
+        K = length(d.cutpoints) + 1
+        @test support(d) == 1:K
+
+        # Adding up probabilities sums to 1
+        s = sum(pdf.(d, support(d)))
+        @test s ≈ 1.0 atol = 0.0001
+    end
+
     @testset "distributions functions" begin
         λ = 0.01:0.01:5
         LLp = @. logpdf(Poisson(λ), 1)