update

fixup

update, tidy

Apply 3 suggestions

Co-authored-by: Miha Zgubic <[email protected]>

add an error

remove error, as closing over `y` breaks inference

simplify, update

solve Core.Box

tests

approx

Author: mcabbott

Project.toml

@@ -14,9 +14,9 @@ Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 
 [compat]
 ChainRulesCore = "1.12"
-ChainRulesTestUtils = "1.5"
-Compat = "3.42.0"
-FiniteDifferences = "0.12.20"
+ChainRulesTestUtils = "1.6"
+Compat = "3.42"
+FiniteDifferences = "0.12.24"
 IrrationalConstants = "0.1.1"
 JuliaInterpreter = "0.8,0.9"
 RealDot = "0.1"
@@ -33,3 +33,4 @@ Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
 [targets]
 test = ["ChainRulesTestUtils", "FiniteDifferences", "JuliaInterpreter", "Random", "StaticArrays", "Test"]
+

src/rulesets/Base/mapreduce.jl

@@ -232,62 +232,51 @@ end
 rrule(::typeof(cumsum), x::AbstractVector) = rrule(cumsum, x; dims=1)
 
 #####
-##### `maximum`, `minimum`
+##### `maximum(f, xs)`, `minimum(f, xs)`
 #####
 
+# Rules for `maximum(x)` live with `findmax(x)` in array.jl
+
 for mimum in (:minimum, :maximum)
-    pullback1 = Symbol(mimum, :_pullback_f)
-    pullback2 = Symbol(mimum, :_pullback_composed)
     findm = Symbol(:find, string(mimum)[1:3])
 
     @eval function rrule(
-        config::RuleConfig{>:HasReverseMode}, ::typeof($mimum), f::F, xs::AbstractArray{<:Number}; dims=:
+        config::RuleConfig{>:HasReverseMode},
+        ::typeof($mimum),
+        f::F,
+        xs::AbstractArray{<:Number};
+        dims=:,
     ) where {F}
         project = ProjectTo(xs)
-
-        # The easy case is when we can use `findmax` to get index, and write into it:
-        if dims isa Colon && VERSION >= v"1.7-"
-            y, ind = $findm(f, xs)
-            function $pullback1(dy)
-                # Notice this evaluates `f` one more time, but this shouldn't matter
-                # unless `f` is sateful, in which case both this and `maximum(f.(xs))` 
-                # give undefined results.
-                _, one_back = rrule_via_ad(config, f, xs[ind])
-                df, one_dx_raw = one_back(unthunk(dy))
-                one_dx = unthunk(one_dx_raw)
-                x_thunk = @thunk project(_writezero(xs, one_dx, ind, dims))
-                x_ithunk = InplaceableThunk(x_thunk) do dxs
-                    view(dxs, ind) .+= one_dx
-                    dxs
-                end
-                return (NoTangent(), df, x_ithunk)
+        if dims isa Colon && VERSION >= v"1.7"
+            # The fast case is when we can use `findmax` to get index, and write into it:
+            y1, ind = $findm(f, xs)  # (Julia 1.6 doesn't have this method.)
+            function minormax_f_back1(dy)
+                # Notice this evaluates `f` one more time, but this shouldn't matter unless `f` is
+                # stateful, in which case both this and `maximum(f.(xs))` give uncertain results.
+                y_ad, one_back = rrule_via_ad(config, f, xs[ind])
+                isapprox(y_ad, y1) || throw(ArgumentError("expected `f` to give same result with AD, got $y_ad != $y1"))
+                df, one_dx = one_back(unthunk(dy))
+                dxs = _zerolike_writeat(xs, unthunk(one_dx), dims, ind) # TODO make _zerolike_writeat handle thunks
+                return (NoTangent(), df, project(dxs))
             end
-            return y, $pullback1
+            return y1, minormax_f_back1
 
-        # Otherwise, the best path is to broadcast, `maximum(f.(xs); dims)`:
         else
-            mid, cast_back = rrule_via_ad(config, broadcast, f, xs; dims=dims)
-            y, max_back = rrule($mimum, fxs; dims=dims)
-            function $pullback2(dys)
-                _, dmid = max_back(dys)
-                _, df, dxs = cast_back(dmid)  # if cast_back from rrule_via_ad makes an InplaceableThunk,
-                return (NoTangent(), df, project(dxs))  # then this project() will give an error.
+            # Otherwise, the best path is to broadcast, `maximum(f.(xs); dims)`:
+            fxs, cast_back = rrule_via_ad(config, broadcast, f, xs)
+            y2, mm_back = rrule($mimum, fxs; dims)
+            function minormax_f_back2(dy)
+                _, dmid = mm_back(dy)
+                _, df, dxs = cast_back(dmid)
+                return (NoTangent(), df, project(dxs))
             end
-            return y, $pullback2
-        end
+            return y2, minormax_f_back2
 
+        end
     end # @eval function rrule(...)
 end
 
-# from another PR:
- function _writezero(x, dy, ind, dims)
-     # It's unfortunate to close over `x`, but `similar(typeof(x), axes(x))` doesn't 
-     # allow `eltype(dy)`, nor does it work for many structured matrices.
-     dx = fill!(similar(x, eltype(dy), axes(x)), false)
-     view(dx, ind) .= dy  # possibly 0-dim view, allows dy::Number and dy::Array, and dx::CuArray
-     dx
- end
-
 #####
 ##### `prod`
 #####

test/rulesets/Base/mapreduce.jl

@@ -2,8 +2,6 @@
 Base.sum(xs::AbstractArray, weights::AbstractArray) = dot(xs, weights)
 struct SumRuleConfig <: RuleConfig{Union{HasReverseMode}} end
 
-const CFG = ChainRulesTestUtils.ADviaRuleConfig()
-
 @testset "Reductions" begin
     @testset "sum(::Tuple)" begin
         test_frule(sum, Tuple(rand(5)))
@@ -137,23 +135,22 @@ const CFG = ChainRulesTestUtils.ADviaRuleConfig()
     end
 
     @testset "maximum(f, xs)" begin
-        # This calls back into AD
-        test_rrule(maximum, abs, [-4.0, 2.0, 2.0], check_inferred=false)
-        test_rrule(minimum, sqrt, Float64[1 2; 3 4], check_inferred=false)
-        test_rrule(maximum, Multiplier(2.0), [2.0, 4.0, 8.0], check_inferred=false)  # Multiplier defined in test_helpers.jl
-
-        # dims keyword
-        @test_skip test_rrule(maximum, sqrt, Float64[1 2; 3 4], fkwargs=(;dims=1), check_inferred=false)
-        @test_skip test_rrule(minimum, abs, randn(3,3), fkwargs=(;dims=2), check_inferred=false)
+        test_rrule(maximum, abs, [-4.0, 2.0, 2.0])
+        test_rrule(minimum, sqrt, Float64[1 2; 3 4])
+        test_rrule(maximum, Multiplier(2.0), [2.0, 4.0, 8.0])  # Multiplier defined in test_helpers.jl
 
         # repeated -- can't use FiniteDifferences
-        y1, bk1 = rrule(TestConfigReverse(), maximum, abs, [-4.0, 2.0, 4.0, 2.0])  # TestConfigReverse defined in test_helpers.jl
+        y1, bk1 = rrule(CFG, maximum, abs, [-4.0, 2.0, 4.0, 2.0])  # CFG defined in test_helpers.jl
         @test y1 === 4.0
-        @test unthunk(bk1(10.0)[3]) == [-10, 0, 0, 0]
+        @test unthunk(bk1(10.0)[3]) ≈ [-10, 0, 0, 0]
+
+        # dims keyword -- these call `rrule_via_ad(broadcast, ...`
+        test_rrule(maximum, sqrt, Float64[1 2; 3 4], fkwargs=(; dims = 1), check_inferred=false)
+        test_rrule(minimum, abs, randn(3,3), fkwargs=(; dims = 2), check_inferred=false)
 
-        # y2, bk2 = rrule(TestConfigReverse(), minimum, abs, [1 2 3; -5 -4 -4], dims=2)
-        # @test y2 == hcat([1, 4])
-        # @test unthunk(bk2(hcat([10, 20]))[3]) == [10 0 0; 0 -20 0]
+        y2, bk2 = rrule(CFG, minimum, abs, [1 2 3; -5 -4 -4], dims = 2)
+        @test y2 == hcat([1, 4])
+        @test_broken unthunk(bk2(hcat([10, 20]))[3]) ≈ [10 0 0; 0 -20 0]  # This used to work? Fine in Zygote
     end
 
     @testset "prod" begin

test/test_helpers.jl

@@ -1,3 +1,6 @@
+
+const CFG = ChainRulesTestUtils.TestConfig()  # CRTU v1.6
+
 """
     Multiplier(x)
 
@@ -97,25 +100,6 @@ function ChainRulesCore.rrule(::typeof(make_two_vec), x)
     return make_two_vec(x), make_two_vec_pullback
 end
 
-# Trivial rule configurations, allowing `rrule_via_ad` with simple functions:
-struct TestConfigReverse <: RuleConfig{HasReverseMode} end
-function ChainRulesCore.rrule_via_ad(::TestConfigReverse, args...; kw...)
-    if hasmethod(rrule, typeof(args), keys(kw))
-        rrule(args...; kw...)
-    else
-        error("TestConfigReverse can only handle `rrule_via_ad(f, args...)` when there is an rrule method")
-    end
-end
-
-struct TestConfigForwards <: RuleConfig{HasForwardsMode} end
-function ChainRulesCore.frule_via_ad(::TestConfigReverse, args...; kw...)
-    if hasmethod(frule, typeof(args), keys(kw))
-        frule(args...; kw...)
-    else
-        error("TestConfigForwards can only handle `frule_via_ad(f, args...)` when there is an frule method")
-    end
-end
-
 @testset "test_helpers.jl" begin
 
     @testset "Multiplier" begin