Merge pull request #51 from JuliaNLSolvers/pkm/optimbase

Remove Optim as dependency and add OptimBase instead.

.travis.yml

@@ -1,7 +1,7 @@
 language: julia
 julia:
-    - 0.4
     - 0.5
+    - 0.6
     - nightly
 notifications:
     email: false

REQUIRE

@@ -1,5 +1,5 @@
-julia 0.4
+julia 0.5
 Calculus
 Distributions
-Optim 0.7.5
 Compat 0.9.1
+OptimBase

src/LsqFit.jl

@@ -4,10 +4,12 @@ module LsqFit
            estimate_errors,
            estimate_covar
 
-    using Optim
     using Calculus
     using Distributions
     using Compat
+    using OptimBase
+
+    import Base.summary
 
     include("levenberg_marquardt.jl")
     include("curve_fit.jl")

src/curve_fit.jl

@@ -11,10 +11,10 @@ end
 # provide a method for those who have their own Jacobian function
 function lmfit(f::Function, g::Function, p0, wt; kwargs...)
     results = levenberg_marquardt(f, g, p0; kwargs...)
-    p = Optim.minimizer(results)
+    p = minimizer(results)
     resid = f(p)
     dof = length(resid) - length(p)
-    return LsqFitResult(dof, p, f(p), g(p), Optim.converged(results), wt)
+    return LsqFitResult(dof, p, f(p), g(p), converged(results), wt)
 end
 
 function lmfit(f::Function, p0, wt; kwargs...)

src/levenberg_marquardt.jl

@@ -1,5 +1,5 @@
-immutable LevenbergMarquardt <: Optim.Optimizer end
-
+immutable LevenbergMarquardt <: Optimizer end
+Base.summary(::LevenbergMarquardt) = "Levenberg-Marquardt"
 """
     `levenberg_marquardt(f, g, initial_x; <keyword arguments>`
 
@@ -69,10 +69,10 @@ function levenberg_marquardt{T}(f::Function, g::Function, initial_x::AbstractVec
     n_buffer = Vector{T}(n)
 
     # Maintain a trace of the system.
-    tr = Optim.OptimizationTrace{LevenbergMarquardt}()
+    tr = OptimizationTrace{LevenbergMarquardt}()
     if show_trace
         d = Dict("lambda" => lambda)
-        os = Optim.OptimizationState{LevenbergMarquardt}(iterCt, sum(abs2, fcur), NaN, d)
+        os = OptimizationState{LevenbergMarquardt}(iterCt, sum(abs2, fcur), NaN, d)
         push!(tr, os)
         println(os)
     end
@@ -145,7 +145,7 @@ function levenberg_marquardt{T}(f::Function, g::Function, initial_x::AbstractVec
         if show_trace
             g_norm = norm(J' * fcur, Inf)
             d = Dict("g(x)" => g_norm, "dx" => delta_x, "lambda" => lambda)
-            os = Optim.OptimizationState{LevenbergMarquardt}(iterCt, sum(abs2, fcur), g_norm, d)
+            os = OptimizationState{LevenbergMarquardt}(iterCt, sum(abs2, fcur), g_norm, d)
             push!(tr, os)
             println(os)
         end
@@ -161,19 +161,22 @@ function levenberg_marquardt{T}(f::Function, g::Function, initial_x::AbstractVec
         converged = g_converged | x_converged
     end
 
-    Optim.MultivariateOptimizationResults(
-        "Levenberg-Marquardt", # method
+    MultivariateOptimizationResults(
+        LevenbergMarquardt(),    # method
         initial_x,             # initial_x
         x,                     # minimizer
         sum(abs2, fcur),       # minimum
         iterCt,                # iterations
         !converged,            # iteration_converged
         x_converged,           # x_converged
         0.0,                   # x_tol
+        0.0,
         false,                 # f_converged
         0.0,                   # f_tol
+        0.0,
         g_converged,           # g_converged
         tolG,                  # g_tol
+        0.0,
         false,                 # f_increased
         tr,                    # trace
         f_calls,               # f_calls

test/levenberg_marquardt.jl

@@ -9,7 +9,7 @@ let
     initial_x = [100.0, 100.0]
 
     results = LsqFit.levenberg_marquardt(f_lm, g_lm, initial_x)
-    @assert norm(Optim.minimizer(results) - [0.0, 2.0]) < 0.01
+    @assert norm(OptimBase.minimizer(results) - [0.0, 2.0]) < 0.01
 
 
     function rosenbrock_res(x, r)
@@ -36,12 +36,12 @@ let
 
     results = LsqFit.levenberg_marquardt(frb, grb, initial_xrb)
 
-    @assert norm(Optim.minimizer(results) - [1.0, 1.0]) < 0.01
+    @assert norm(OptimBase.minimizer(results) - [1.0, 1.0]) < 0.01
 
     # check estimate is within the bound PR #278
      result = LsqFit.levenberg_marquardt(frb, grb, [150.0, 150.0]; lower = [10.0, 10.0], upper = [200.0, 200.0])
-     @test Optim.minimizer(result)[1] >= 10.0
-     @test Optim.minimizer(result)[2] >= 10.0
+     @test OptimBase.minimizer(result)[1] >= 10.0
+     @test OptimBase.minimizer(result)[2] >= 10.0
 
 
 
@@ -60,7 +60,7 @@ let
         g_lsq = Calculus.jacobian(f_lsq)
         results = LsqFit.levenberg_marquardt(f_lsq, g_lsq, [0.5, 0.5])
 
-        @assert norm(Optim.minimizer(results) - [1.0, 2.0]) < 0.05
+        @assert norm(OptimBase.minimizer(results) - [1.0, 2.0]) < 0.05
     end
 
     let
@@ -84,15 +84,15 @@ let
 
         lower=[5.0, 11.0, 5.0]
         results = LsqFit.levenberg_marquardt(f_lsq, g_lsq, [15.0, 15.0, 15.0], lower=lower)
-        Optim.minimizer(results)
-        @test Optim.converged(results)
-        @test all(Optim.minimizer(results) .>= lower)
+        OptimBase.minimizer(results)
+        @test OptimBase.converged(results)
+        @test all(OptimBase.minimizer(results) .>= lower)
 
         upper=[15.0, 9.0, 15.0]
         results = LsqFit.levenberg_marquardt(f_lsq, g_lsq, [5.0, 5.0, 5.0], upper=upper)
-        Optim.minimizer(results)
-        @test Optim.converged(results)
-        @test all(Optim.minimizer(results) .<= upper)
+        OptimBase.minimizer(results)
+        @test OptimBase.converged(results)
+        @test all(OptimBase.minimizer(results) .<= upper)
 
         # tests for PR #267
         LsqFit.levenberg_marquardt(f_lsq, g_lsq, [15.0, 15.0, 15.0], show_trace=true)

test/runtests.jl

@@ -2,7 +2,7 @@
 # Correctness Tests
 #
 
-using LsqFit, Optim, Base.Test, Compat
+using LsqFit, Base.Test, Compat
 
 my_tests = [ "curve_fit.jl", "levenberg_marquardt.jl"]