Calculate derivatives using Jacobian-vector products

Author: devmotion

.github/workflows/ci.yml

@@ -20,7 +20,6 @@ jobs:
         with:
           version: ${{ matrix.julia-version }}
       - uses: julia-actions/cache@v2
-      - uses: julia-actions/julia-buildpkg@v1
       - uses: julia-actions/julia-runtest@v1
         with:
           annotate: true

Project.toml

@@ -14,9 +14,8 @@ DoubleFloats = "1"
 ExplicitImports = "1.14"
 JET = "0.9, 0.10"
 LinearAlgebra = "<0.0.1, 1"
-NLSolversBase = "7"
+NLSolversBase = "8"
 NaNMath = "1"
-Optim = "1"
 OptimTestProblems = "2"
 Printf = "<0.0.1, 1"
 Test = "<0.0.1, 1"
@@ -27,9 +26,11 @@ Aqua = "4c88cf16-eb10-579e-8560-4a9242c79595"
 DoubleFloats = "497a8b3b-efae-58df-a0af-a86822472b78"
 ExplicitImports = "7d51a73a-1435-4ff3-83d9-f097790105c7"
 JET = "c3a54625-cd67-489e-a8e7-0a5a0ff4e31b"
-Optim = "429524aa-4258-5aef-a3af-852621145aeb"
 OptimTestProblems = "cec144fc-5a64-5bc6-99fb-dde8f63e154c"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
 [targets]
-test = ["Aqua", "ExplicitImports", "JET", "Test", "OptimTestProblems", "Optim", "DoubleFloats"]
+test = ["Aqua", "ExplicitImports", "JET", "Test", "OptimTestProblems", "DoubleFloats"]
+
+[sources]
+NLSolversBase = { url = "https://github.com/devmotion/NLSolversBase.jl.git", rev = "dmw/jvp" }

src/LineSearches.jl

@@ -1,7 +1,7 @@
 module LineSearches
 
 using Printf: @sprintf
-using LinearAlgebra: dot, norm
+using LinearAlgebra: norm
 using NaNMath: NaNMath
 using NLSolversBase: NLSolversBase, AbstractObjective
 
@@ -28,11 +28,8 @@ function make_ϕdϕ(df, x_new, x, s)
         # Move a distance of alpha in the direction of s
         x_new .= x .+ α.*s
 
-        # Evaluate ∇f(x+α*s)
-        NLSolversBase.value_gradient!(df, x_new)
-
         # Calculate ϕ(a_i), ϕ'(a_i)
-        NLSolversBase.value(df), real(dot(NLSolversBase.gradient(df), s))
+        return NLSolversBase.value_jvp!(df, x_new, s)
     end
     ϕdϕ
 end
@@ -41,49 +38,16 @@ function make_ϕ_dϕ(df, x_new, x, s)
         # Move a distance of alpha in the direction of s
         x_new .= x .+ α.*s
 
-        # Evaluate ∇f(x+α*s)
-        NLSolversBase.gradient!(df, x_new)
-
         # Calculate ϕ'(a_i)
-        real(dot(NLSolversBase.gradient(df), s))
+        return NLSolversBase.jvp!(df, x_new, s)
     end
     make_ϕ(df, x_new, x, s), dϕ
 end
 function make_ϕ_dϕ_ϕdϕ(df, x_new, x, s)
-    function dϕ(α)
-        # Move a distance of alpha in the direction of s
-        x_new .= x .+ α.*s
-
-        # Evaluate f(x+α*s) and ∇f(x+α*s)
-        NLSolversBase.gradient!(df, x_new)
-
-        # Calculate ϕ'(a_i)
-        real(dot(NLSolversBase.gradient(df), s))
-    end
-    function ϕdϕ(α)
-        # Move a distance of alpha in the direction of s
-        x_new .= x .+ α.*s
-
-        # Evaluate ∇f(x+α*s)
-        NLSolversBase.value_gradient!(df, x_new)
-
-        # Calculate ϕ'(a_i)
-        NLSolversBase.value(df), real(dot(NLSolversBase.gradient(df), s))
-    end
-    make_ϕ(df, x_new, x, s), dϕ, ϕdϕ
+    make_ϕ_dϕ(df, x_new, x, s)..., make_ϕdϕ(df, x_new, x, s)
 end
 function make_ϕ_ϕdϕ(df, x_new, x, s)
-    function ϕdϕ(α)
-        # Move a distance of alpha in the direction of s
-        x_new .= x .+ α.*s
-
-        # Evaluate ∇f(x+α*s)
-        NLSolversBase.value_gradient!(df, x_new)
-
-        # Calculate ϕ'(a_i)
-        NLSolversBase.value(df), real(dot(NLSolversBase.gradient(df), s))
-    end
-    make_ϕ(df, x_new, x, s), ϕdϕ
+    make_ϕ(df, x_new, x, s), make_ϕdϕ(df, x_new, x, s)
 end
 
 include("types.jl")

test/examples.jl

@@ -14,7 +14,6 @@ my_tests = [
     "initial.jl",
     "alphacalc.jl",
     "arbitrary_precision.jl",
-    "examples.jl",
     "captured.jl",
     "issues.jl",
     "qa.jl",