feldtruby 0.4.4 → 0.4.5

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA1:
-  metadata.gz: f017a746cfb0c942893d7207ddcd0ec3d3a2a252
-  data.tar.gz: 475222949c01be1ae9dbb5678b4f2da00bf5a877
+  metadata.gz: b5558f3b77d1f52c3bb73f77ab79755b9d8e8f20
+  data.tar.gz: 1d6e7c4cea54211d5487b41b71710ea30827ceb8
 SHA512:
-  metadata.gz: b85a1f419930ec29d072227bfa3423b3088f2befc4755d60df2cf8f870407d35cb84b2684074917c38aa3904e697476bc6ea9d5ecb633e0692e9a2c274fb6c02
-  data.tar.gz: f825a6ee4384822a133f39df4f8c1867f42e611e2cdb307bcf2eaf3771860435f2ed1cd7440210b0841c2040f83e4affd1744023350154e40095dd272679b6d1
+  metadata.gz: d7f2ff3e1edcacd55278c11d228eced97c128ce1628f55f5edb30aee3fdfb6c2a7c676ac62860a62e5cfd33bd588eba795d50f489becbaae592bb63b36f8b07b
+  data.tar.gz: 2397c74ee5a6eea1c250b9481ace101ed6ecc58990649b2fa6d10aaee73136a5b5e758ccd7689c807f1283f49493c7e3ac1987dea11b37fe4fe262f6e8474020

data/lib/feldtruby/logger.rb

@@ -9,7 +9,7 @@ module FeldtRuby
 class Logger
   DefaultParams = {
     :verbose => false,
-    :printFrequency => 0.4  # Minimum seconds between consecutive messages printed for the same event type
+    :printFrequency => 0.3  # Minimum seconds between consecutive messages printed for the same event type
   }
 
   UnixEpoch = Time.at(0)

data/lib/feldtruby/optimize/differential_evolution.rb

@@ -67,7 +67,7 @@ class DEOptimizerBase < EvolutionaryOptimizer
 				"Trial Quality" => @objective.quality_of(trial),
 				"Target" => target, 
 				"Target Quality" => @objective.quality_of(target)
-				}, "DE (step #{@num_optimization_steps}): Trial vector was better than target vector", true
+				}, "DE (step #{@num_optimization_steps}): Trial vector was better than target vector"
 			update_candidate_in_population(target_index, trial)
 			feedback_on_trial_vs_target(trial, target, true)
 		else
@@ -142,49 +142,58 @@ module DE_CrossoverStrategy_Binomial
 	end
 end
 
-# Building block for mutation strategies.
-module DE_X_1_StrategyBuildingBlock
-	# We need 0 target parents and 2 other parents. Note that we must sample a target parent
-	# also even though it is not used in the mutation.
-	def num_parents_to_sample; 3; end
-
-	def difference_vector(donorParentsIndices)
-		p1, p2 = get_candidates_with_indices(donorParentsIndices)
-		(p1 - p2)
-	end
-end
+# The most-used DE/rand/1/* mutation strategy.
+module DE_MutationStrategy_Rand_1
+	# We need three parents for donor vector. And then the target, so 1+3 in total.
+	def num_parents_to_sample; 4; end
 
-module DE_Rand_X_StrategyBuildingBlock
 	def mutate(targetIndex, donorParentsIndices)
-		p3 = get_candidate(donorParentsIndices[-1])
-		p3 + scale_factor(targetIndex) * difference_vector(donorParentsIndices[0...-1])
+		p1, p2, p3 = get_candidates_with_indices(donorParentsIndices)
+		p3 + (scale_factor(targetIndex) * (p1 - p2))
 	end
 end
 
-# The most-used DE/rand/1 mutation strategy.
-module DE_MutationStrategy_Rand_1
-	include DE_X_1_StrategyBuildingBlock
+# DE/rand/1/bin uses 
+#   Bounding  = random bounding within the search space
+#   Update 	  = no updates based on feedback
+#   Crossover = Classic binomial
+#   Mutation  = Rand-1
+class DEOptimizer_Rand_1_Bin < DEOptimizerBase
+	include DE_BoundingStrategy_RandomWithinSearchSpace
+	include DE_UpdateStrategy_NoFeedbackUpdates
+	include DE_CrossoverStrategy_Binomial
+	include DE_MutationStrategy_Rand_1
+end
 
-	# We need one more parent in the Rand strategy than in the others, but
-	# we can reuse the difference vector generation. So partial reuse here
-	# NOTE! Order of inclusion is critical!!!
-	def num_parents_to_sample; 4; end
+# The DE/best/1/* mutation strategy.
+module DE_MutationStrategy_Best_1
+	# We need two parents for donor vector. And then the target, so 1+2 in total.
+	def num_parents_to_sample; 3; end
 
-	include DE_Rand_X_StrategyBuildingBlock
+	def mutate(targetIndex, donorParentsIndices)
+		p1, p2 = get_candidates_with_indices(donorParentsIndices)
+		candidate_from_array(best) + (scale_factor(targetIndex) * (p1 - p2))
+	end
 end
 
-# The default DEOptimizer uses 
-#   Bounding  = random bouding within the search space
+# DE/best/1/bin uses
+#   Bounding  = random bounding within the search space
 #   Update 	  = no updates based on feedback
 #   Crossover = Classic binomial
-#   Mutation  = Rand-1
-class DEOptimizer < DEOptimizerBase
+#   Mutation  = Best-1
+class DEOptimizer_Best_1_Bin < DEOptimizerBase
 	include DE_BoundingStrategy_RandomWithinSearchSpace
 	include DE_UpdateStrategy_NoFeedbackUpdates
 	include DE_CrossoverStrategy_Binomial
-	include DE_MutationStrategy_Rand_1
+	include DE_MutationStrategy_Best_1
 end
 
+# DE/rand/1/bin is the default DE optimizer since it does not converge too
+# quickly but is generally good. For many problems the DEOptimizer_Best_1_Bin
+# gives better results faster though.
+DEOptimizer = DEOptimizer_Rand_1_Bin
+#DEOptimizer = DEOptimizer_Best_1_Bin
+
 # Optimize the _numVariables_ between the _min_ and _max_ values given _costFunction_.
 # Default is to minimize.
 def self.optimize(min, max, options = {:verbose => true}, 

data/lib/feldtruby/optimize/objective.rb

@@ -294,9 +294,6 @@ class Objective
         :type_of_improvement => typeOfReset
         }, "Better candidate found for goal #{goal_methods[index]}"
 
-      # Reset the best object since we have a new scale
-      @best_candidate = nil
-
     end
 
     inc_version_number

data/lib/feldtruby/version.rb

@@ -1,3 +1,3 @@
 module FeldtRuby
-  VERSION = "0.4.4"
+  VERSION = "0.4.5"
 end

data/test/long_running/single_objective_problems.rb

@@ -169,7 +169,7 @@ end
 class MinEggHolder < Min2DSingleObjectiveFunc
   def minimum
     # -959.6407
-    -963.5808501270542
+    -963.5808501270571
   end
 
   def min_solutions

data/test/long_running/test_single_objective_optimization.rb

@@ -19,10 +19,15 @@ module MiniTest::Expectations
   infect_an_assertion :assert_close_to_one_solution, :must_be_close_to_one_solution_of
 end
 
-def best_from_de_on_objective(objective, dimensions, numSteps = 25_000, verbose = false)
+# Note! We have set the numSteps below so that DEOptimizer_Rand_1_Bin can converge
+# to a solution. However, for most of these optimization problems 
+# DEOptimizer_Best_1_Bin converges much faster, and can thus be used.
+
+def best_from_de_on_objective(objective, dimensions, numSteps = 25_000, 
+  verbose = false, optimizer = FeldtRuby::Optimize::DEOptimizer_Rand_1_Bin)
   objective.dimensions = dimensions if objective.respond_to?(:dimensions=)
   ss = objective.search_space
-  de = DEOptimizer.new(objective, ss, {:verbose => verbose, 
+  de = optimizer.new(objective, ss, {:verbose => verbose, 
     :maxNumSteps => numSteps})
 
   start_time = Time.now
@@ -74,34 +79,17 @@ end
 
 describe "Easom function" do
   it 'can optimize the Easom function' do
-    objective = MinEasom.new
-    ss = objective.search_space
-    de = DEOptimizer.new(objective, ss, {:verbose => false, 
-      :maxNumSteps => 30_000, :printFrequency => 0.0, 
-      :samplerRadius => 5})
-    best = de.optimize().to_a
-
-    val = objective.calc_func(best)
-    val.must_be_close_to objective.minimum
-    val.must_be :>=, objective.minimum
-
-    best.must_be_close_to_one_solution_of objective, 0.01
+    best, obj, time = best_from_de_on_objective MinBeale.new, nil, 10_000, false, FeldtRuby::Optimize::DEOptimizer_Best_1_Bin
+    best.must_be_close_to_one_solution_of obj
+    time.must_be :<, 1.5
   end
 end
 
 describe "EggHolder function" do
   it 'can optimize the Eggholder function' do
-    objective = MinEggHolder.new
-    ss = objective.search_space
-    de = DEOptimizer.new(objective, ss, {:verbose => false, 
-      :maxNumSteps => 25_000, :samplerRadius => 6})
-    best = de.optimize().to_a
-
-    val = objective.calc_func(best)
-    val.must_be_close_to objective.minimum
-    val.must_be :>=, objective.minimum
-
-    best.must_be_close_to_one_solution_of objective, 0.01
+    best, obj, time = best_from_de_on_objective MinEggHolder.new, nil, 10_000, false, FeldtRuby::Optimize::DEOptimizer_Best_1_Bin
+    best.must_be_close_to_one_solution_of obj
+    time.must_be :<, 1.5
   end
 end
 

data/test/test_optimize_differential_evolution.rb

@@ -41,4 +41,26 @@ describe "DifferentialEvolution" do
 		de2.best.sum.must_be :<=, 0.40
 		de2.num_optimization_steps.must_equal 1234
 	end
+end
+
+describe "DE/best/1/bin" do
+	it "works for rms of small vector" do
+		s2 = SearchSpace.new_symmetric(2, 1)
+		o1 = MinimizeRMS.new
+		de1 = DEOptimizer_Best_1_Bin.new(o1, s2, {:verbose => false, :maxNumSteps => 1000})
+		de1.optimize()
+		# Very unlikely we get a number over 0.30 (2 elements) after 1000 steps...
+		de1.best.sum.must_be :<=, 0.30
+		de1.num_optimization_steps.must_equal 1000
+	end
+
+	it "works for rms and sum of small vector and with more steps" do
+		s4 = SearchSpace.new_symmetric(4, 1)
+		o2 = MinimizeRMSAndSum.new
+		de2 = DEOptimizer_Best_1_Bin.new(o2, s4, {:verbose => false, :maxNumSteps => 1234})
+		de2.optimize()
+		# Very unlikely we get a number over 0.40 (4 elements)...
+		de2.best.sum.must_be :<=, 0.40
+		de2.num_optimization_steps.must_equal 1234
+	end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: feldtruby
 version: !ruby/object:Gem::Version
-  version: 0.4.4
+  version: 0.4.5
 platform: ruby
 authors:
 - Robert Feldt