updated the repeated_bboptimize and compare_optimizers since they still used old bboptimize interface; this fixes the ackley_100dim example

Author: robertfeldt

examples/ackley_100dim.jl

@@ -1,11 +1,15 @@
 using BlackBoxOptim
 
-p = BlackBoxOptim.example_problems["Ackley"]
-
-BlackBoxOptim.repeated_bboptimize(5, p, 100, [
+problem = BlackBoxOptim.example_problems["Ackley"]
+numrepeats = 5
+dim = 100
+methods = [
   :generating_set_search, 
   :probabilistic_descent, 
   :adaptive_de_rand_1_bin_radiuslimited,
   :random_search,
-  :separable_nes], 
-  20.0)
+  :separable_nes]
+max_time = 10.0
+ftol = 1e-5
+
+BlackBoxOptim.repeated_bboptimize(numrepeats, problem, dim, methods, max_time, ftol)

src/compare_optimizers.jl

@@ -109,8 +109,10 @@ function report_from_result_dict(statsdict)
   report_on_values("Fitness", statsdict[:fitnesses], "  ")
   report_on_values("Time", statsdict[:times], "  ")
   report_on_values("Num function evals", statsdict[:numevals], "  ")
-  println("  Success rate: ", round(pdict["Within fitness tolerance of optimum"], 3), "%\n")
-  pdict["Within fitness tolerance of optimum"]
+
+  success_rate = round(get(pdict, "Within fitness tolerance of optimum", 0.0), 3)
+  println("  Success rate: ", success_rate, "%\n")
+  success_rate
 end
 
 function rank_result_dicts_by(result_dicts, byfunc, desc; rev = false,
@@ -139,42 +141,41 @@ function report_on_methods_results_on_one_problem(problem, result_dicts, numrepe
   rank_result_dicts_by(result_dicts, (d) -> median(d[:fitnesses]), "fitness"; descsummary = "median")
   rank_result_dicts_by(result_dicts, (d) -> median(d[:times]), "time (in seconds)";
     descsummary = "median", rpad = " secs")
-  rank_result_dicts_by(result_dicts, (d) -> int(median(d[:numevals])), "num function evals"; descsummary = "median")
+  rank_result_dicts_by(result_dicts, (d) -> round(Int, median(d[:numevals])), "num function evals"; descsummary = "median")
 
   for rd in result_dicts
     report_from_result_dict(rd)
   end
 
 end
 
-function repeated_bboptimize(numrepeats, problem, dim, methods, max_time, ftol = 1e-5, parameters::Parameters = EMPTY_PARAMS)
+function repeated_bboptimize(numrepeats, problem, dim, methods, max_time, ftol = 1e-5, extraparams::Parameters = EMPTY_PARAMS)
 
   fp = instantiate(problem, dim)
   result_dicts = ParamsDict[]
 
   # Just so they are declared
   ps = best_so_far = nothing
 
-  params = chain(parameters, ParamsDict(:FitnessTolerance => ftol))
+  params = chain(extraparams, ParamsDict(:FitnessTolerance => ftol))
 
   for m in methods
 
     ts, fs, nes = zeros(numrepeats), zeros(numrepeats), zeros(Int, numrepeats)
-    rcounts = Dict{String,Int}("Within fitness tolerance of optimum" => 0)
+    rcounts = Dict{String,Int}()
 
-    for i in 1:numrepeats
-      p = fp # BlackBoxOptim.ShiftedAndBiasedProblem(fp)
-      best, fs[i], reason, ts[i], ps, nes[i] = bboptimize(p; max_time = max_time,
-        method = m, parameters = params)
-      rcounts[reason] = 1 + get(rcounts, reason, 0)
-    end
+    results = map(1:numrepeats) do i
+      result = bboptimize(fp, params; MaxTime = max_time, Method = m)
 
-    if best_so_far == nothing
-      best_so_far = worst_fitness(ps[:Evaluator])
-    end
+      # Save key data about this run:
+      ts[i]  = elapsed_time(result)
+      fs[i]  = best_fitness(result)
+      nes[i] = f_calls(result)
 
-    # ???
-    best_so_far =
+      # And count only the general stop reasons
+      reason = general_stop_reason(result)
+      rcounts[reason] = 1 + get(rcounts, reason, 0)
+    end
 
     rdict = ParamsDict(:method => m, :fitnesses => fs, :times => ts, :numevals => nes, :reasoncounts => rcounts)
     rdict[:success_rate] = report_from_result_dict(rdict)

test/test_bboptimize.jl

@@ -18,7 +18,7 @@ end
         end
 
         @testset "example #3" begin
-            res = bboptimize(rosenbrock2d; SearchRange = (-5.0, 5.0), NumDimensions = 2, method = :de_rand_1_bin, TraceMode = :silent)
+            res = bboptimize(rosenbrock2d; SearchRange = (-5.0, 5.0), NumDimensions = 2, Method = :de_rand_1_bin, TraceMode = :silent)
             @test best_fitness(res) < 0.001
         end
 

test/test_parameters.jl

@@ -85,6 +85,14 @@
             dc = chain(d3, d1, d2)
             @test (dc[:a], dc[:b], dc[:c]) == (2, 4, 5)
         end
+
+        @testset "Check chaining from real use case" begin
+            # This sequence is used in compare_optimizers.jl:repeated_bboptimize
+            parameters = BlackBoxOptim.EMPTY_PARAMS
+            ftol = 1e-5
+            params = chain(parameters, ParamsDict(:FitnessTolerance => ftol))
+            @test params[:FitnessTolerance] == ftol
+        end
     end
 
     @testset "converting to Dict" begin