Collect traces, and add ParticleSwarm to the mix.

src/Optim.jl

@@ -85,7 +85,7 @@ module Optim
     # Heuristic Optimization Methods
     include("nelder_mead.jl")
     #include("simulated_annealing.jl")
-    #include("particle_swarm.jl")
+    include("particle_swarm.jl")
 
     # Univariate methods
     include("golden_section.jl")

src/bfgs.jl

@@ -11,26 +11,6 @@ method_string(method::BFGS) = "BFGS"
 BFGS(; linesearch!::Function = hz_linesearch!, initial_invH = x -> eye(eltype(x), length(x))) =
   BFGS(linesearch!, initial_invH)
 
-function trace!(tr, state, iteration, method::BFGS, options)
-    dt = Dict()
-    if options.extended_trace
-        dt["x"] = copy(state.x)
-        dt["g(x)"] = copy(state.g)
-        dt["~inv(H)"] = copy(state.invH)
-        dt["Current step size"] = state.alpha
-    end
-    g_norm = vecnorm(state.g, Inf)
-    update!(tr,
-            iteration,
-            state.f_x,
-            g_norm,
-            dt,
-            options.store_trace,
-            options.show_trace,
-            options.show_every,
-            options.callback)
-end
-
 type BFGSState{T}
     n::Int64
     x::Array{T}

src/nelder_mead.jl

@@ -82,23 +82,6 @@ centroid(simplex, h) = centroid!(similar(simplex[1]), simplex, h)
 
 nmobjective(y::Vector, m::Integer, n::Integer) = sqrt(var(y) * (m / n))
 
-function trace!(tr, state, iteration, method::NelderMead, options)
-    dt = Dict()
-    if options.extended_trace
-        dt["centroid"] = state.x_centroid
-        dt["step_type"] = state.step_type
-    end
-    update!(tr,
-            iteration,
-            state.f_lowest,
-            state.f_x,
-            dt,
-            options.store_trace,
-            options.show_trace,
-            options.show_every,
-            options.callback)
-end
-
 function print_header(method::NelderMead)
     @printf "Iter     Function value    √(Σ(yᵢ-ȳ)²)/n \n"
     @printf "------   --------------    --------------\n"
@@ -147,7 +130,9 @@ type NelderMeadState{T, N}
     f_calls::Int64
     g_calls::Int64
 end
+
 initialize_state(method::NelderMead, options, d, initial_x::Array) = initialize_state(method, options, d.f, initial_x)
+
 function initialize_state{T}(method::NelderMead, options, f::Function, initial_x::Array{T})
     m = length(initial_x)
     n = m + 1

src/optimize.jl

@@ -205,7 +205,7 @@ end
 
 # Iterative method boiler plate
 #This will just be replaced by ::Optimizer once they're all converted
-typealias Refactored Union{AcceleratedGradientDescent, GradientDescent, MomentumGradientDescent, ConjugateGradient, LBFGS, BFGS, NelderMead}
+typealias Refactored Union{AcceleratedGradientDescent, GradientDescent, MomentumGradientDescent, ConjugateGradient, LBFGS, BFGS, NelderMead, ParticleSwarm}
 
 function after_while!(d, state, method, options)
     nothing
@@ -226,6 +226,8 @@ function optimize{T}(d, initial_x::Array{T}, method::Refactored, options::Optimi
     x_converged, f_converged = false, false
     g_converged = if typeof(method) <: NelderMead
         nmobjective(state.f_simplex, state.m, state.n) < options.g_tol
+        elseif  typeof(method) <: ParticleSwarm
+            g_converged = false
         else
             vecnorm(state.g, Inf) < options.g_tol
         end