Merge pull request #20 from JuliaAI/dev

For a 0.1.3 release

Author: ablaom

Project.toml

@@ -1,7 +1,7 @@
 name = "CategoricalDistributions"
 uuid = "af321ab8-2d2e-40a6-b165-3d674595d28e"
 authors = ["Anthony D. Blaom <[email protected]>"]
-version = "0.1.2"
+version = "0.1.3"
 
 [deps]
 CategoricalArrays = "324d7699-5711-5eae-9e2f-1d82baa6b597"

src/arrays.jl

@@ -243,6 +243,12 @@ function Base.Broadcast.broadcasted(::typeof(mode),
     mode_flat = map(1:length(u)) do i
         max_prob = maximum(dic[ref][i] for ref in keys(dic))
         m = zero(R)
+        
+        # `maximum` of any iterable containing `NaN` would return `NaN` 
+        # For this case the index `m` won't be updated in the loop as relations
+        # involving NaN as one of it's argument always returns false 
+        # (e.g `==(NaN, NaN)` returns false)
+        throw_nan_error_if_needed(max_prob)
         for ref in keys(dic)
             if dic[ref][i] == max_prob
                 m = ref

src/methods.jl

@@ -196,6 +196,12 @@ function Dist.mode(d::UnivariateFinite)
     p = values(dic)
     max_prob = maximum(p)
     m = first(first(dic)) # mode, just some ref for now
+
+    # `maximum` of any iterable containing `NaN` would return `NaN` 
+    # For this case the index `m` won't be updated in the loop below as relations
+    # involving NaN as one of it's arguments always returns false
+    # (e.g `==(NaN, NaN)` returns false)
+    throw_nan_error_if_needed(max_prob)
     for (x, prob) in dic
         if prob == max_prob
             m = x
@@ -205,6 +211,18 @@ function Dist.mode(d::UnivariateFinite)
     return d.decoder(m)
 end
 
+function throw_nan_error_if_needed(x)
+    if isnan(x)
+        throw(
+            DomainError(
+                NaN,
+                "`mode` is invalid for `UnivariateFininite` distribution "*
+                "with `pdf` containing `NaN`s"
+            )
+        )
+    end
+end
+
 # mode(v::Vector{UnivariateFinite}) = mode.(v)
 # mode(u::UnivariateFiniteVector{2}) =
 #     [u.support[ifelse(s > 0.5, 2, 1)] for s in u.scores]

src/utilities.jl

@@ -21,7 +21,8 @@ always true.
 
 Not to be confused with `levels(x.pool)`. See the example below.
 
-Also, overloaded for `x` a `CategoricalArray` or `CategoricalPool`.
+Also, overloaded for `x` a `CategoricalArray`, `CategoricalPool`, and
+for views of `CategoricalArray`.
 
 **Private method.*
 
@@ -57,7 +58,7 @@ Also, overloaded for `x` a `CategoricalArray` or `CategoricalPool`.
 classes(p::CategoricalPool) = [p[i] for i in 1:length(p)]
 classes(x::CategoricalValue) = classes(CategoricalArrays.pool(x))
 classes(v::CategoricalArray) = classes(CategoricalArrays.pool(v))
-
+classes(v::SubArray{<:Any, <:Any, <:CategoricalArray}) = classes(parent(v))
 
 # # CATEGORICAL VALUES TO INTEGERS
 

test/arrays.jl

@@ -198,6 +198,17 @@ end
     u   = UnivariateFinite(P, pool=missing)
     expected = mode.([u...])
     @test all(mode.(u) .== expected)
+
+    # `mode` broadcasting of `Univariate` objects containing `NaN` in probs.
+    unf_arr = UnivariateFinite(
+        [
+            0.1 0.2 NaN 0.1 NaN;
+            0.2 0.1 0.1 0.4 0.2;
+            0.3 NaN 0.2 NaN 0.3
+        ],
+        pool=missing
+    )
+    @test_throws DomainError mode.(unf_arr)
 end
 
 @testset "cat for UnivariateFiniteArray" begin

test/methods.jl

@@ -175,6 +175,10 @@ end
     p[42] = 0.5
     d = UnivariateFinite(v, p)
     @test mode(d) == 42
+
+    # `mode` of `Univariate` objects containing `NaN` in probs.
+    unf = UnivariateFinite([0.1, 0.2, NaN, 0.1, NaN], pool=missing)
+    @test_throws DomainError mode(unf)
 end
 
 @testset "UnivariateFinite methods" begin

test/types.jl

@@ -36,7 +36,7 @@ end
     UnivariateFinite(supp, probs, pool=missing, augment=true);
 
     # dimension mismatches:
-    badprobs = rand(40, 3)
+    badprobs = rand(rng, 40, 3)
     @test_throws(CategoricalDistributions.err_dim(supp, badprobs),
                  UnivariateFinite(supp, badprobs, pool=missing))
 
@@ -60,7 +60,7 @@ end
     probs = probs ./ sum(probs)
     u = UnivariateFinite(probs, pool=missing);
     @test u isa UnivariateFinite
-    probs = rand(10, 2)
+    probs = rand(rng, 10, 2)
     probs = probs ./ sum(probs, dims=2)
     u = UnivariateFinite(probs, pool=missing);
     @test u.scitype == Multiclass{2}
@@ -73,9 +73,11 @@ end
 
     v = categorical(1:3)
     @test_logs((:warn, r"Ignoring"),
-               UnivariateFinite(v[1:2], rand(3), augment=true, pool=missing));
+               UnivariateFinite(v[1:2], rand(rng, 3),
+                                augment=true, pool=missing));
     @test_logs((:warn, r"Ignoring"),
-               UnivariateFinite(v[1:2], rand(3), augment=true, ordered=true));
+               UnivariateFinite(v[1:2], rand(rng, 3),
+                                augment=true, ordered=true));
 
     # using `UnivariateFiniteArray` as a constructor just falls back
     # to `UnivariateFinite` constructor:

test/utilities.jl

@@ -17,6 +17,8 @@ import CategoricalDistributions: classes, transform, decoder, int
     levels!(v, reverse(levels(v)))
     @test classes(v[1]) == levels(v)
     @test classes(v) == levels(v)
+    vsub = view(v, 1:2)
+    @test classes(vsub) == classes(v)
 end
 
 @testset "int, classes, decoder" begin