charlie 0.7.1 → 0.8.0

data/History.txt

@@ -1,3 +1,13 @@
+== 0.8.0 / 2008-02-12
+* Added BlendingCrossover
+* Benchmark setup/teardown blocks
+* Made population inherit array
+* Support for GP-style evolution (never apply both crossover and mutation)
+* Co-evolution with small tournaments, iterated prisoner's dilemma example.
+* Added evolve_multiple* methods
+* Added support for matrix genotypes, 2-d arrays of floats or bits.
+* Added a genotype for neural networks
+
 == 0.7.1 / 2008-01-12
 * Fixed a major bug in RouletteSelection and ScaledRouletteSelection.
 

data/Manifest.txt

@@ -13,7 +13,8 @@ examples/EXAMPLES_README.txt
 examples/bitstring.rb
 examples/coevolution.rb
 examples/function_optimization.rb
-examples/output/bitstring_royalroad.html
+examples/matrix.rb
+examples/neural.rb
 examples/output/function_optimization_sombrero.html
 examples/output/function_optimization_twopeak.csv
 examples/output/function_optimization_twopeak.html
@@ -33,6 +34,8 @@ lib/charlie/genotype.rb
 lib/charlie/list/list_crossover.rb
 lib/charlie/list/list_genotype.rb
 lib/charlie/list/list_mutate.rb
+lib/charlie/list/matrix.rb
+lib/charlie/list/neural.rb
 lib/charlie/mutate.rb
 lib/charlie/permutation/permutation.rb
 lib/charlie/population.rb
@@ -42,7 +45,9 @@ test/t_common.rb
 test/test_benchmark.rb
 test/test_cross.rb
 test/test_evolve.rb
+test/test_matrix.rb
 test/test_mutator.rb
+test/test_neural.rb
 test/test_permutation.rb
 test/test_sel.rb
 test/test_tree.rb

data/README.txt

@@ -1,4 +1,4 @@
-== Charlie version 0.7.1
+== Charlie version 0.8.0
 * http://rubyforge.org/projects/charlie/
 * http://charlie.rubyforge.org
 * mailto:sander.land+ruby@gmail.com

data/Rakefile

@@ -19,7 +19,7 @@ h = Hoe.new('charlie', Charlie::VERSION) do |p|
   p.changes = p.paragraphs_of('History.txt', 0..1).join("\n\n")
 
   #p.test_globs ['test/test_all.rb']  # removed test_all, because rake now does that anyway
-  p.clean_globs = ['test/output/*','./**/*~'] # Remove this on "rake clean" : test output and kate backups
+  p.clean_globs = ['test/output/*','./**/*~','./**/*.rbc'] # Remove this on "rake clean" : test output and kate backups, rubinius stuff
 
   p.rdoc_pattern = /^lib|\.txt$|^data\/[A-Z]+(\..+)?$/  # include data/* doc files
   p.remote_rdoc_dir = '' # Release to root

data/TODO.txt

@@ -1,19 +1,14 @@
 This is my todo list. Contributions and suggestions don't have to be restricted to these things in any way.
 
 === Small stuff
-* Arithmetic crossovers, Brood crossover? 
+* Brood crossover? 
 === Bigger stuff
 * More builtin genotypes:
   * Tree-based, extending into genetic programming. mostly done. add option for ADFs?
-  * Graphs / adjacency matrix. Multidimensional arrays in general?
-  * Neural networks.
-* Method for population to do several shorter runs until convergence (cf. gridwalk example).
-* Method for population to do several shorter runs, save some the best in each to generate a new population, and evolve this. Is this useful?
-* More selection algorithms.  Multiple populations with migration? Niching?  gladiatorial/tournaments (+Prisoner's dilemma example)?
+* More selection algorithms: multiple populations with migration? Niching?
 * Extend benchmarking options:
   * Extend dsl to have options for manual test cases given by sel/cross/mut/etc options, or even free code blocks.
   * Option to compare several generation/population size settings. Especially both of them together while keeping the number of fitness evaluations roughly constant.
-  * Instead of (2), constraint on time, or # of fitness evals instead of generations. (?)
   * Detailed stats for use in comparing a small number of strategies? Including population stats for every generation and run, graphs?
 
 

data/data/BENCHMARK

@@ -46,6 +46,10 @@ complete, but will improve the accuracy of the results.
 * Default is the fitness itself.
 * This can be an array, if you want to track multiple stats at once (for example: tracking both training and generalization error).
 
+==== <tt>setup{|population| ... }</tt> and <tt>teardown{|population| ... }</tt>
+* Extra blocks called before and after the tests (before Population#evolve_silent, and after 'track_stats')
+* setup can be used to reshuffle traning data, initialize the population to some specific state, etc.
+
 ==== Remarks 
 * If you want to test the module that is already included without repeating it, just pass Module.new as one of the arguments (or Module.new{self.name='default'} for something more descriptive). This is also the default, so <tt>GABenchmark.benchmark(klass){}</tt> will simply test the strategies already included.
 * The methods <tt>selection, crossover, mutator, generations, population_size, repeat</tt> are defined using metaprogramming as <tt>attr_dsl</tt>. Passing no arguments to the function will return the current value, passing more will assign the value. This means you  can do things like <tt>selection << TournamentSelection(3)</tt> in the block. 

data/data/CROSSOVER

@@ -20,7 +20,7 @@ Using the Genotype.use keyword does this automatically.
 === NullCrossover
 Just returns copies of the two parents, i.e. performs no crossover at all.
 
-== List-based Crossovers
+== List-based Crossovers  <tt>->file</tt>[link:files/lib/charlie/list/list_crossover_rb.html]
 These crossovers can be used for all list- and string-based genotypes.
 
 === SinglePointCrossover
@@ -32,6 +32,9 @@ Standard uniform crossover.Returns two children.
 === NPointCrossover(n=2)
 N-point crossover. Returns two children.
 
+=== BlendingCrossover(exploration_alpha=0.1,type=:cube)
+Blending crossover as used in ES
+
 
 == Specialized Crossovers
 === For PermutationGenotype
@@ -41,10 +44,15 @@ It is fairly destructive, which can lead to poor performance.
 EdgeRecombinationCrossover is a less destructive crossover for permutations.
 It's reasonably effective, but fairly slow and only works for permutations of 0...n, so you have to look up the actual elements in the fitness function.
 
+=== For FloatMatrixGenotype, BitMatrixGenotype
+* MatrixUniformCrossover does a uniform crossover (per-element) on matrices.
+* The normal SinglePointCrossover, UniformCrossover, NPointCrossover can also be used. They act on rows as single elements.
+
+
 === For TreeGenotype
 TreeCrossover does a standard subtree swapping crossover for trees. Returns two children.
 
-== Meta-crossovers
+== Meta-crossovers  <tt>->file</tt>[link:files/lib/charlie/crossover_rb.html]
 These functions take one or more crossover modules and generate a new crossover module.
 
 === SingleChild(crossover)  (#SingleChild)

data/data/GENOTYPE

@@ -11,7 +11,7 @@ The base class includes NullCrossover, NullMutator and Elitism(ScaledRouletteSel
 All of the other genotypes described here inherit the base class, and its associated selection/crossover/mutation operators, unless
 otherwise indicated.
 
-=== List-based genotypes  (file)[link:files/lib/charlie/list/list_genotype_rb.html]
+=== List-based genotypes  <tt>->file</tt>[link:files/lib/charlie/list/list_genotype_rb.html]
 All of these genotypes are based on arrays and can use the list-based crossover and mutation operators.
 They are all generated dynamically by functions, and should be used like:
 
@@ -37,13 +37,33 @@ Genotype of +n+ elements (not necessarily chars).
 
 Includes <tt>ListMutator(:expected_n[2],:replace[*elements]), SinglePointCrossover</tt> by default.
 
+==== NeuralNetworkGenotype(input_n,hidden_n,output_n=1,scaling=1.0,hidden_f=NN_TANH,output_f=NN_TANH)
+Genotype for a neural network. <tt>->file</tt>[link:files/lib/charlie/list/neural_rb.html]
+
+Inherits FloatListGenotype and includes <tt>UniformCrossover</tt> by default.
+
+
+=== Matrix-based genotypes  <tt>->file</tt>[link:files/lib/charlie/list/matrix_rb.html]
+These genotypes are based on arrays of arrays and can use the matrix-based crossover and mutation operators. Some of the crossovers for lists also work on these, by considering rows as single elements of a list (see the CROSSOVER file for details).
+
+====   FloatMatrixGenotype(rows,columns=rows,range=0..1)   
+Genotype of <tt>rows x columns</tt> floats initialized in the range +range+. Mutations can take the floats out of this range.
+
+Includes <tt>MatrixMutator()</tt> (with default parameters :expected_n[3], :uniform[ 0.25 ]) and <tt>MatrixUniformCrossover</tt> by default.
+
+====  BitStringGenotype(n)
+Genotype of <tt>rows x columns</tt> bits.
+
+Includes <tt>MatrixMutator(:expected_n[3],:flip)</tt> and <tt>MatrixUniformCrossover</tt> by default.
+
+
 === Specialized genotypes
 These genotypes have their own crossover and mutation operators.
 
-==== PermutationGenotype(n,elements=0...n)  (file)[link:files/lib/charlie/permutation/permutation_rb.html]
+==== PermutationGenotype(n,elements=0...n)  <tt>->file</tt>[link:files/lib/charlie/permutation/permutation_rb.html]
 Genotype for permutations. Includes TranspositionMutator and PermutationCrossover by default.
 
-==== TreeGenotype(terminals, unary_ops, binary_ops)  (file)[link:files/lib/charlie/tree/tree_rb.html]
+==== TreeGenotype(terminals, unary_ops, binary_ops)  <tt>->file</tt>[link:files/lib/charlie/tree/tree_rb.html]
 * Genotype for trees. Useful for genetic programming. Includes PMutate(0.5,TreeReplaceMutator) and PCross(0.7,TreeCrossover) by default.
 * Functions included: eval_genes(terminal_values_hash), size, depth, module GPTreeHelper
 * Check examples/tree.rb for usage.

data/data/MUTATION

@@ -18,7 +18,7 @@ The non-destructive counterpart +mutate+ is defined automatically by Genotype#mu
 Just returns self
 
 == List-based Mutators
-These crossovers can be used for all list- and string-based genotypes.
+These mutators can be used for all list- and string-based genotypes.
 
 === ListMutator(strategy=:expected_n ,point_mutator=:uniform)
 Generates a wide variety of mutators for lists, strings and bitstrings.
@@ -29,16 +29,28 @@ See the documentation of list_mutate.rb[link:files/lib/charlie/list/list_mutate_
 
 Both of these parameters use Symbol#[] in the examples. <tt>:replace['a','b']</tt> is simply equivalent to <tt>[:replace,'a','b']</tt>
 
+== Matrix-based Mutators
+These mutators can be used for all matrix genotypes.
+
+=== MatrixMutator(strategy=:expected_n ,point_mutator=:uniform)
+Generates a wide variety of mutators for matrices of floats or bits.
+
+* strategy can be a proc or one of the MatrixMutationStrategies : single point, multi-point and probabilistic. See the documentation of matrix.rb[link:files/lib/charlie/list/matrix_rb.html] for more info on these.
+* point_mutator can be a proc or one of the PointMutators, as with ListMutator.
+
+
 == Specialized Mutators
+
+
 === For PermutationGenotype
-All of these will probably work on other array/string-based genotypes as well, but this is untested.
+All of these will probably work on other array/string/matrix-based genotypes as well, but this is untested.
 
 * TranspositionMutator is a transposition mutator for permutations. It interchanges two elements and leaves the remaining elements in their original positions.
 * InversionMutator is another mutator for permutations. It inverses a part of the genes array.
 * InsertionMutator takes a random element of the permutation, and inserts it at a random position.
 * RotationMutator takes a random element, and rotates the permutation such that this is the first element, i.e. it effectively does nothing if the permutation represents a cycle.
 
-=== For TreeGenotype  (file)[link:files/lib/charlie/tree/tree_rb.html]
+=== For TreeGenotype  <tt>->file</tt>[link:files/lib/charlie/tree/tree_rb.html]
 * TreeReplaceMutator(depth,type) replaces a randomly chosen subtree with a new, randomly generated, subtree.
 * TreePruneMutator replaces a randomly chosen subtree with a randomly generated terminal.
 * TreeRemoveNodeMutator replaces a random node by one of its children.

data/data/SELECTION

@@ -20,13 +20,13 @@ Using the Genotype.use keyword does this automatically.
 These strategies work with genotype classes which define a +fitness+ function. The +fitness+ function should return a number, higher fitness means better and more likely to be selected. All except RouletteSelection can handle negative fitness values.
 
 *   RandomSelection
-*   TruncationSelection(best=0.3)
-*   BestOnlySelection
+*   TruncationSelection(best=0.3), TruncationSelection = TruncationSelection()
+*   BestOnlySelection = TruncationSelection(1)
 *   RouletteSelection
-*   ScaledRouletteSelection(&block)
-*   TournamentSelection(group_size=4,n_times=nil)   
+*   ScaledRouletteSelection(&block), ScaledRouletteSelection = RankSelection = ScaledRouletteSelection()
+*   TournamentSelection(group_size=4,n_times=nil), TournamentSelection = TournamentSelection()
 
-The documentation for selection_rb[link:files/lib/charlie/selection_rb.html] contains explanations for most of these.
+The documentation for <tt>selection.rb</tt>[link:files/lib/charlie/selection_rb.html] contains explanations for most of these.
 
 === Meta-selection
 ==== Elitism(selection,n=1)
@@ -36,13 +36,20 @@ Generates a module which applies elitism to some selection strategy. This:
 * replaces the last +n+ individuals in the new population with the best +n+ of the old population (unless the newer individual has higher fitness).
 This ensures the maximum fitness can never decrease.
 
+==== GP(selection,crossover_probability=0.75)
+Applies crossover OR mutation instead of both for generating children. Common in genetic programming. Only works for selection methods which use yield to generate children (currently all of the built-in selection methods do this).
 
 == Co-evolutionary selection strategies
-These strategies work with genotype classes which define a <tt>fight(other)</tt> function. 
-This function should return true if the instance defeats +other+ in a 'fight'.
 
 *  GladiatorialSelection
+   * This strategy works with genotype classes which define a <tt>fight(other)</tt> function.  This function should return true if the instance defeats +other+ in a 'fight'.
+
+*  CoTournamentSelection(group_size=4,full_tournament=false,n_times=nil)
+   * This strategy works with genotype classes which define a <tt>fight_points(other)</tt> function.  This function should return [points for self,points for other] as a result from a 'fight'. Points should always be >= 0.
+   * Selection is total-points-proportional (as in roulette selection) within tournaments of size <tt>group_size</tt>.
 
 Because there is no way to determine who is the actual best individual, using Population#benchmark 
 for comparing strategies for convergence speed, etc.
-is not possible. Defining a fitness function as some statistic you want to track and then using benchmark should be possible, but is untested.
+is not possible. Defining a fitness function as some statistic you want to track or using <tt>track_stats</tt> to override the default and then using benchmark should be possible, but is untested.
+
+

data/examples/EXAMPLES_README.txt

@@ -1,70 +1,91 @@
-In this directory are several examples of how to use the library.
+In this directory (<tt>/examples</tt>) are several examples of how to use the library.
 This file serves as an index, which can be used to find an example for a specific function.
 
-====  function_optimization.rb  ====
+===  function_optimization.rb
 - ruby function_optimization.rb hill
-Description: Function optimization example in many dimensions, hill climbing.
-Uses: FloatListGenotype, TournamentSelection, UniformCrossover, ListMutator
+  * Description: Function optimization example in many dimensions, hill climbing.
+  * Uses: FloatListGenotype, TournamentSelection, UniformCrossover, ListMutator
  
 - ruby function_optimization.rb twopeak
-Description: Function optimization example in 1 dimension, avoiding a local maximum.
-Uses: FloatListGenotype (single float), NullCrossover, GABenchmark
+  * Description: Function optimization example in 1 dimension, avoiding a local maximum.
+  * Uses: FloatListGenotype (single float), NullCrossover, GABenchmark
 
 - ruby function_optimization.rb sombrero
-Description: Function optimization example in 2 dimensions, avoiding many maxima.
-Uses: FloatListGenotype, GABenchmark
+  * Description: Function optimization example in 2 dimensions, avoiding many maxima.
+  * Uses: FloatListGenotype, GABenchmark
 
-====  bitstring.rb  ====
+=== bitstring.rb
 - ruby bitstring.rb 512
-Description: Very simple example that finds the binary representation of 512.
-Uses: BitStringGenotype
+  * Description: Very simple example that finds the binary representation of 512.
+  * Uses: BitStringGenotype
  
 - ruby bitstring.rb royalroad
-Description: Example for convergence with limited feedback from the fitness function.
-Uses: BitStringGenotype, TruncationSelection, UniformCrossover, ListMutator, cache_fitness, GABenchmark
+  * Description: Example for convergence with limited feedback from the fitness function.
+  * Uses: BitStringGenotype, TruncationSelection, UniformCrossover, ListMutator, cache_fitness, GABenchmark
 
-====  string.rb  ====
+===  string.rb
 - ruby string.rb weasel
-Description: Simple target string evolver.
-Uses: StringGenotype, BestOnlySelection, UniformCrossover, ListMutator, GABenchmark
+  * Description: Simple target string evolver.
+  * Uses: StringGenotype, BestOnlySelection, UniformCrossover, ListMutator, GABenchmark
  
 - ruby string.rb gridwalk
-Description: Example of using strings of symbols other than chars to evolve a route through a grid.
-Uses: StringGenotype, TournamentSelection, evolve_until
+  * Description: Example of using strings of symbols other than chars to evolve a route through a grid.
+  * Uses: StringGenotype, TournamentSelection, evolve_until
 
-==== tsp.rb  ====
+=== tsp.rb
 - ruby tsp.rb
-Description: Travelling salesperson problem.
-Uses: PermutationGenotype, PCross, PMutateN, evolve_until, TournamentSelection,
-      EdgeRecombinationCrossover, InversionMutator ,InsertionMutator, GABenchmark
+  * Description: Travelling salesperson problem.
+  * Uses: PermutationGenotype, PCross, PMutateN, evolve_until, TournamentSelection,
+  *       EdgeRecombinationCrossover, InversionMutator ,InsertionMutator, GABenchmark
 
-====  permutation.rb  ====
+===  permutation.rb 
 - ruby permutation.rb
-Description: send+more=money problem with PermutationGenotype.
-Uses: PermutationGenotype
+  * Description: send+more=money problem with PermutationGenotype.
+  * Uses: PermutationGenotype
 
-====  coevolution.rb  ====
+===  coevolution.rb 
 - ruby coevolution.rb simple
-Description: Simple co-evolutionary example. Evolves a two-element array to be [large,small]
-Uses: GladiatorialSelection, FloatListGenotype, UniformCrossover, ListMutator
+  * Description: Simple co-evolutionary example. Evolves a two-element array to be [large,small]
+  * Uses: GladiatorialSelection, FloatListGenotype, UniformCrossover, ListMutator
 
 - ruby coevolution.rb sunburn
-Description: More elaborate example of co-evolution. Evolves weapon systems on space ships (like Ashlock (2004), ch.4).
-Uses: GladiatorialSelection, StringGenotype, custom mutation/crossover, UniformCrossover
+  * Description: More elaborate example of co-evolution. Evolves weapon systems on space ships (like Ashlock (2004), ch.4).
+  * Uses: GladiatorialSelection, StringGenotype, custom mutation/crossover, UniformCrossover
 
-====  tree.rb  ====
+- ruby coevolution.rb prisoner
+  * Description: Iterated prisoner's dilemma.
+  * Uses: CoTournamentSelection
+
+===  tree.rb 
 - ruby tree.rb cos
-Description: Example for TreeGenotype/GP. Evolves a polynomial that approximates cos(x).
-Uses: TreeGenotype, ...
+  * Description: Example for TreeGenotype/GP. Evolves a polynomial that approximates cos(x).
+  * Uses: TreeGenotype, ...
 
 - ruby tree.rb pors [n=31]
-Description: Plus One Recall Store example. Generate a number n using a minimal # of +,1,STO,RCL operations.
-Uses: TreeGenotype, passing proc{}'s to node values at evaluation, custom unary operators.
+  * Description: Plus One Recall Store example. Generate a number n using a minimal # of +,1,STO,RCL operations.
+  * Uses: TreeGenotype, passing proc{}'s to node values at evaluation, custom unary operators.
 
 - ruby tree.rb porsx [n=31]
-Description: Same idea as PORS, but generate a tree that evaluates to n*T where T can only be recalled once.
+  * Description: Same idea as PORS, but generate a tree that evaluates to n*T where T can only be recalled once.
+
+- ruby tree.rb xor
+  * Description: Finds the smallest tree for A XOR B using |,&,!.
 
 - ruby tree.rb bloat
-Description: Just grows large trees.
+  * Description: Just grows large trees.
+
+===  matrix.rb 
+- ruby matrix.rb royalmatrix
+  * Description: Like bitstring.rb royalroad.
+  * Uses: BitMatrixGenotype
+
+- ruby matrix.rb inverse
+  * Description: Finds the inverse of a matrix.
+  * Uses: FloatMatrixGenotype
+
+===  neural.rb 
+- ruby neural.rb
+  * Description: Finds a neural network for the XOR function.
+  * Uses: NeuralNetworkGenotype
 
 

data/examples/coevolution.rb

@@ -7,7 +7,7 @@ end
 
 if ARGV[0].nil?
   puts "Several examples for Co-evolution."
-  puts "Usage: ruby coevolution.rb simple|sunburn"
+  puts "Usage: ruby coevolution.rb simple|sunburn|prisoner"
 
 
 
@@ -144,4 +144,45 @@ puts "Results:",pop
 
 
 
+elsif ARGV[0].downcase.gsub(/[^a-z]/,'') == 'prisoner'
+
+
+
+
+puts "Runs a tournament-based co-evolutionary genetic algorithm of the Prisoner's Dilemma. "
+
+class Prisoner <  BitStringGenotype(5)
+  RESULT = {'00'=>1,'01'=>5,'10'=>0,'11'=>3}
+#look at last ft, 00 01 10 11 -> ix in genes .. genes[4] = init
+# rev for other! 3-x?
+  def fight_points(other) # fight_points(other) return [points for self,points for other]
+    strat, other_strat = genes, other.genes
+    move, other_move = strat[4], other_strat[4] # initial move
+    s, os = 0,0
+    100.times{
+      my_res = "#{move}#{other_move}"
+      other_res = my_res.reverse
+      s  += RESULT[my_res]
+      os += RESULT[other_res]
+      move, other_move = strat[my_res.to_i(2)], other_strat[other_res.to_i(2)]
+    }
+    [s,os]
+  end
+  BIT = ['D','C']
+  NAMES = { # name some strategies
+    "1111"=>"ALL-C","0000"=>"ALL-D","0101"=>"TIT-FOR-TAT","0001"=>"GRUDGE (ALL-D-AFTER-ANY-D)","1001"=>"PAVLOV",
+    "0011"=>"ALL-START","0100"=>"ALL-C-AFTER-ANY-C","0111"=>"ALL-C-IF-START-C"
+  }
+  def to_s
+    long_name = "DD => #{BIT[genes[0]]} CD => #{BIT[genes[2]]} | DC => #{BIT[genes[1]]}  CC=>#{BIT[genes[3]]}  | START #{BIT[genes[4]]}"
+    "#{long_name} \t #{NAMES[super[0..3]]}"
+  end
+  use CoTournamentSelection(5), UniformCrossover, ListMutator(:expected_n[0.25],:flip)
+end
+
+pop = Population.new(Prisoner,50).evolve_silent(200)
+puts "PREVIOUS MOVE OF SELF/OTHER => ACTION \t\t\t NAME"
+puts pop
+
+
 end

data/examples/matrix.rb

@@ -0,0 +1,49 @@
+begin
+  require '../lib/charlie'
+rescue LoadError
+  require 'rubygems'
+  require 'charlie'
+end
+require 'pp'
+
+if ARGV[0]=='royalmatrix'
+
+puts "Like bitstring.rb royalroad, but with a matrix."
+
+class RRMatrix < BitMatrixGenotype(16,4)
+  def fitness
+    genes.count{|row| row.all?{|b| b==1 } }
+  end
+  def to_s ; genes.map{|row| row.all?{|b| b==1 } ? '*':'_' }.join ; end
+  use MatrixMutator(:expected_n_per_row[0.1],:flip), UniformCrossover # row-based uniform crossover
+end
+r = Population.new(RRMatrix).evolve_on_console(200)
+
+elsif ARGV[0]=='inverse'
+
+puts "Evolves the inverse of a matrix. Not very effective."
+
+N = 3
+M = Array.new(N){Array.new(N){rand}}
+I = (0...N).map{|i|  (0...N).map{|j| j==i ? 1 : 0 }}
+
+class MInverse < FloatMatrixGenotype(N,N,-1..1)
+  def fitness
+    a = genes; b = M
+    diff = 0
+    (0...N).each{|i|  (0...N).each{|j|
+      diff += (I[i][j] - (0...N).map{|k| a[i][k] * b[k][j] }.sum).abs
+    }}
+    -diff
+  end
+  def to_s
+    "\n" + genes.map{|r| r.map{|e| "%.2f" % e }.join("\t") }.join("\n")
+  end
+  use MatrixMutator(:expected_n[3],:gaussian[0.1])
+end
+r = Population.new(MInverse).evolve_on_console(200)
+
+else
+  puts "Several examples for matrices."
+  puts "Usage: ruby matrix.rb royalmatrix|inverse"
+end