Merge pull request #141 from control-toolbox/differentiationinterface

WIP: Differentiationinterface

Project.toml

@@ -5,6 +5,7 @@ version = "0.9.1"
 
 [deps]
 DataStructures = "864edb3b-99cc-5e75-8d2d-829cb0a9cfe8"
+DifferentiationInterface = "a0c0ee7d-e4b9-4e03-894e-1c5f64a51d63"
 DocStringExtensions = "ffbed154-4ef7-542d-bbb7-c09d3a79fcae"
 ForwardDiff = "f6369f11-7733-5829-9624-2563aa707210"
 Interpolations = "a98d9a8b-a2ab-59e6-89dd-64a1c18fca59"
@@ -21,6 +22,7 @@ Unicode = "4ec0a83e-493e-50e2-b9ac-8f72acf5a8f5"
 
 [compat]
 DataStructures = "0.18"
+DifferentiationInterface = "0.5"
 DocStringExtensions = "0.9"
 ForwardDiff = "0.10"
 Interpolations = "0.15"

src/CTBase.jl

@@ -14,7 +14,8 @@ module CTBase
 # using
 import Base
 using DocStringExtensions
-using ForwardDiff: jacobian, gradient, ForwardDiff # automatic differentiation
+using DifferentiationInterface: AutoForwardDiff, derivative, gradient, jacobian, prepare_derivative, prepare_gradient, prepare_jacobian
+import ForwardDiff
 using Interpolations: linear_interpolation, Line, Interpolations # for default interpolation
 using MLStyle # pattern matching
 using Parameters # @with_kw: to have default values in struct
@@ -193,6 +194,8 @@ See also: [`ctVector`](@ref), [`DState`](@ref).
 """
 const DCostate   = ctVector
 
+__auto() = AutoForwardDiff() # default AD backend
+
 #
 include("exception.jl")
 include("description.jl")

src/print.jl

@@ -190,7 +190,7 @@ function Base.show(io::IO, ::MIME"text/plain", ocp::OptimalControlModel{<: TimeD
     #
     println(io)
     printstyled(io, "Declarations ", bold=true)
-    printstyled(io, "(* required):\n", bold=false)
+    printstyled(io, "(* for required):\n", bold=false)
     #println(io)
 
     # print table of settings

src/utils.jl

@@ -73,17 +73,19 @@ $(TYPEDSIGNATURES)
 
 Return the gradient of `f` at `x`.
 """
-function ctgradient(f::Function, x::ctNumber)
-    return ForwardDiff.derivative(x -> f(x), x)
+function ctgradient(f::Function, x::ctNumber; backend=__auto())
+    extras = prepare_derivative(f, backend, x)
+    return derivative(f, backend, x, extras)
 end
 
 """
 $(TYPEDSIGNATURES)
 
 Return the gradient of `f` at `x`.
 """
-function ctgradient(f::Function, x)
-    return ForwardDiff.gradient(f, x)
+function ctgradient(f::Function, x; backend=__auto())
+    extras = prepare_gradient(f, backend, x)
+    return gradient(f, backend, x, extras)
 end
 
 """
@@ -98,16 +100,22 @@ $(TYPEDSIGNATURES)
 
 Return the Jacobian of `f` at `x`.
 """
-function ctjacobian(f::Function, x::ctNumber) 
-    return ForwardDiff.jacobian(x -> f(x[1]), [x])
+function ctjacobian(f::Function, x::ctNumber; backend=__auto())
+    f_number_to_number = only ∘ f ∘ only
+    extras = prepare_derivative(f_number_to_number, backend, x)
+    der = derivative(f_number_to_number, backend, x, extras)
+    return [der;;]
 end
 
 """
 $(TYPEDSIGNATURES)
 
 Return the Jacobian of `f` at `x`.
 """
-ctjacobian(f::Function, x) = ForwardDiff.jacobian(f, x)
+function ctjacobian(f::Function, x; backend=__auto())
+    extras = prepare_jacobian(f, backend, x)
+    return jacobian(f, backend, x, extras)
+end
 
 """
 $(TYPEDSIGNATURES)