public String DescribeShortGeneralized(Individual ind, IEvolutionState state, int subpopulation, int threadnum)
{
if (Start == null)
{
Start = new bool[MIN_ARRAY_SIZE];
Accept = new bool[MIN_ARRAY_SIZE];
Reading = new int[MIN_ARRAY_SIZE];
From = new int[MIN_ARRAY_SIZE];
To = new int[MIN_ARRAY_SIZE];
State1 = new bool[MIN_ARRAY_SIZE];
State2 = new bool[MIN_ARRAY_SIZE];
Reading1 = TensorFactory.Create <int>(MIN_ARRAY_SIZE, MIN_ARRAY_SIZE); // new int[MIN_ARRAY_SIZE][MIN_ARRAY_SIZE];
Reading0 = TensorFactory.Create <int>(MIN_ARRAY_SIZE, MIN_ARRAY_SIZE); // new int[MIN_ARRAY_SIZE][MIN_ARRAY_SIZE];
Epsilon = TensorFactory.Create <int>(MIN_ARRAY_SIZE, MIN_ARRAY_SIZE); // new int[MIN_ARRAY_SIZE][MIN_ARRAY_SIZE];
Reading1L = new int[MIN_ARRAY_SIZE];
Reading0L = new int[MIN_ARRAY_SIZE];
EpsilonL = new int[MIN_ARRAY_SIZE];
}
FullTest(state, ind, threadnum, PosA, NegA);
return(": " +
((double)_totpos) / PosA.Length + " " +
((double)_totneg) / NegA.Length + " " +
(((double)(_totpos + _totneg)) / (PosA.Length + NegA.Length)) + " " +
(((((double)_totpos) / PosA.Length) + (((double)_totneg) / NegA.Length)) / 2) + " " +
Math.Min((((double)_totpos) / PosA.Length), (((double)_totneg) / NegA.Length)) + " : ");
}
public void setExpectedMaxSize(int numRows, int numCols)
{
this.numCols = numCols;
this.numRows = numRows;
minLength = Math.Min(numCols, numRows);
int maxLength = Math.Max(numCols, numRows);
if (dataQR == null || dataQR.Length < numCols || dataQR[0].Length < numRows * 2)
{
//dataQR = new double[ numCols ][ numRows*2 ];
dataQR = TensorFactory.Create <double>(numCols, numRows * 2);
v = new double[maxLength * 2];
gammas = new double[minLength];
}
if (v.Length < maxLength * 2)
{
v = new double[maxLength * 2];
}
if (gammas.Length < minLength)
{
gammas = new double[minLength];
}
}
public override void Setup(IEvolutionState state, IParameter paramBase)
{
base.Setup(state, paramBase);
state.Output.ExitIfErrors();
IParameter kval = new Parameter(EvolutionState.P_EVALUATOR).Push(P_PROBLEM).Push(P_PROBLEMNAME)
.Push(P_KVALUE);
k = state.Parameters.GetInt(kval, null, 0);
// System.out.println("K = " + k);
for (int i = 0; i < _indices.Length; i++)
{
_indices[i] = -1;
}
_indices['A' - 'A'] = 0;
_indices['B' - 'A'] = 1;
_indices['X' - 'A'] = 2;
_indices['Y' - 'A'] = 3;
_indices['Z' - 'A'] = 4;
_indices['W' - 'A'] = 5;
// now do some initialization
IMersenneTwister r = state.Random[0];
_nodeScore = new double[6];
_edgeScore = TensorFactory.Create <double>(2, 6);
for (int i = 0; i < 6; i++)
{
_nodeScore[i] = 2 * r.NextDouble() - 1;
}
// We need to assure that the best fitness is positive (to normalize it to 1)
// A method to do this is to have at least one terminal symbol with a positive score.
bool ok = false;
for (int i = 2; i < 6; i++)
{
if (_nodeScore[i] > 0)
{
ok = true;
}
}
if (!ok)
{
_nodeScore[2] = r.NextDouble();
}
for (int i = 0; i < 2; i++)
{
for (int j = 0; j < 6; j++)
{
_edgeScore[i][j] = r.NextDouble();
}
}
_bestFitness = ComputeBestFitness();
}
public override void Describe(IEvolutionState state, Individual ind, int subpopulation, int threadnum, int log)
{
if (Start == null)
{
Start = new bool[MIN_ARRAY_SIZE];
Accept = new bool[MIN_ARRAY_SIZE];
Reading = new int[MIN_ARRAY_SIZE];
From = new int[MIN_ARRAY_SIZE];
To = new int[MIN_ARRAY_SIZE];
State1 = new bool[MIN_ARRAY_SIZE];
State2 = new bool[MIN_ARRAY_SIZE];
Reading1 = TensorFactory.Create <int>(MIN_ARRAY_SIZE, MIN_ARRAY_SIZE); // new int[MIN_ARRAY_SIZE][MIN_ARRAY_SIZE];
Reading0 = TensorFactory.Create <int>(MIN_ARRAY_SIZE, MIN_ARRAY_SIZE); // new int[MIN_ARRAY_SIZE][MIN_ARRAY_SIZE];
Epsilon = TensorFactory.Create <int>(MIN_ARRAY_SIZE, MIN_ARRAY_SIZE); // new int[MIN_ARRAY_SIZE][MIN_ARRAY_SIZE];
Reading1L = new int[MIN_ARRAY_SIZE];
Reading0L = new int[MIN_ARRAY_SIZE];
EpsilonL = new int[MIN_ARRAY_SIZE];
}
if (Generalize)
{
FullTest(state, ind, threadnum, PosA, NegA);
}
else
{
FullTest(state, ind, threadnum, PosT, NegT);
}
if (Generalize)
{
state.Output.PrintLn("\n\nBest Individual's Generalization Score...\n" +
"Pos: " + _totpos + "/" + PosA.Length +
" Neg: " + _totneg + "/" + NegA.Length +
"\n(pos+neg)/(allpos+allneg): " +
(float)
(((double)(_totpos + _totneg)) / (PosA.Length + NegA.Length)) +
"\n((pos/allpos)+(neg/allneg))/2: " +
(float)
(((((double)_totpos) / PosA.Length) + (((double)_totneg) / NegA.Length)) / 2) +
"\nMin(pos/allpos,neg/allneg): " +
(float)Math.Min((((double)_totpos) / PosA.Length), (((double)_totneg) / NegA.Length)),
log);
}
state.Output.PrintLn("\nBest Individual's NFA\n=====================\n",
log);
state.Output.PrintLn(PrintCurrentNFA(), log);
}
public void Evaluate(IEvolutionState state, Individual ind, int subpop, int threadnum)
{
if (Start == null)
{
Start = new bool[MIN_ARRAY_SIZE];
Accept = new bool[MIN_ARRAY_SIZE];
Reading = new int[MIN_ARRAY_SIZE];
From = new int[MIN_ARRAY_SIZE];
To = new int[MIN_ARRAY_SIZE];
State1 = new bool[MIN_ARRAY_SIZE];
State2 = new bool[MIN_ARRAY_SIZE];
Reading1 = TensorFactory.Create <int>(MIN_ARRAY_SIZE, MIN_ARRAY_SIZE); // new int[MIN_ARRAY_SIZE][MIN_ARRAY_SIZE];
Reading0 = TensorFactory.Create <int>(MIN_ARRAY_SIZE, MIN_ARRAY_SIZE); // new int[MIN_ARRAY_SIZE][MIN_ARRAY_SIZE];
Epsilon = TensorFactory.Create <int>(MIN_ARRAY_SIZE, MIN_ARRAY_SIZE); // new int[MIN_ARRAY_SIZE][MIN_ARRAY_SIZE];
Reading1L = new int[MIN_ARRAY_SIZE];
Reading0L = new int[MIN_ARRAY_SIZE];
EpsilonL = new int[MIN_ARRAY_SIZE];
}
if (!ind.Evaluated) // don't bother reevaluating
{
EdgeData input = (EdgeData)Input;
FullTest(state, ind, threadnum, PosT, NegT);
// the fitness better be KozaFitness!
var f = (KozaFitness)ind.Fitness;
// this is an awful fitness metric, but it's the standard
// one used for these problems. :-(
f.SetStandardizedFitness(state, (float)
(1.0 - (double)(_totpos + _totneg) /
(PosT.Length + NegT.Length)));
// here are two other more reasonable fitness metrics
//f.SetStandardizedFitness(state,(float)
//(1.0 - Math.Min(((double)_totpos)/PosT.Length,
//((double)_totneg)/NegT.Length)));
//f.SetStandardizedFitness(state,(float)
//(1.0 - (((double)_totpos)/PosT.Length +
//((double)_totneg)/NegT.Length)/2.0));
f.Hits = _totpos + _totneg;
ind.Evaluated = true;
}
}
public int RunCartPole(NEATNetwork net, IEvolutionState state)
{
// double in[] = new double[5]; //Input loading array
double out1;
double out2;
double twelve_degrees = 0.2094384;
_x = _xDot = _theta = _thetaDot = 0.0;
_steps = 0;
double[][] input = TensorFactory.Create <double>(1, 5);
while (_steps++ < MAX_STEPS)
{
/*-- setup the input layer based on the four inputs and bias --*/
//setup_input(net,x,x_dot,theta,theta_dot);
input[0][0] = 1.0; //Bias
input[0][1] = (_x + 2.4) / 4.8;
input[0][2] = (_xDot + .75) / 1.5;
input[0][3] = (_theta + twelve_degrees) / .41;
input[0][4] = (_thetaDot + 1.0) / 2.0;
double[] output = GetNetOutput(net, input, state);
/*-- decide which way to push via which output unit is greater --*/
if (output[0] > output[1])
{
_y = 0;
}
else
{
_y = 1;
}
/*--- Apply action to the simulated cart-pole ---*/
cart_pole(_y);
/*--- Check for failure. If so, return steps ---*/
if (_x < -2.4 || _x > 2.4 || _theta < -twelve_degrees || _theta > twelve_degrees)
{
return(_steps);
}
}
return(_steps);
}
/// <summary>
/// Build an NxN rotation matrix[row][column] with a given seed.
/// </summary>
public static double[] /* row */ [] /* column */ BuildRotationMatrix(IEvolutionState state, long rotationSeed, int N)
{
if (rotationSeed == ROTATION_SEED)
{
state.Output.WarnOnce("Default rotation seed being used (" + rotationSeed + ")");
}
IMersenneTwister rand = new MersenneTwisterFast(ROTATION_SEED);
for (int i = 0; i < 624 * 4; i++) // prime the MT for 4 full sample iterations to get it warmed up
{
rand.NextInt();
}
double[] /* row */ [] /* column */ o = TensorFactory.Create <double>(N, N);
// make random values
for (var i = 0; i < N; i++)
{
for (var k = 0; k < N; k++)
{
o[i][k] = rand.NextGaussian();
}
}
// build random values
for (var i = 0; i < N; i++)
{
// extract o[i] -> no
var no = new double[N];
for (var k = 0; k < N; k++)
{
no[k] = o[i][k];
}
// go through o[i] and o[j], modifying no
for (var j = 0; j < i; j++)
{
var d = Dot(o[i], o[j]);
var val = ScalarMul(d, o[j]);
no = Sub(no, val);
}
o[i] = Normalize(no);
}
return(o);
}
public override Object Clone()
{
KLandscapes tmp = (KLandscapes)base.Clone();
tmp._nodeScore = new double[6];
tmp._edgeScore = TensorFactory.Create <double>(2, 6);
tmp._bestFitness = _bestFitness;
tmp.k = k;
for (int i = 0; i < 6; i++)
{
tmp._nodeScore[i] = _nodeScore[i];
for (int j = 0; j < 2; j++)
{
tmp._edgeScore[j][i] = _edgeScore[j][i];
}
}
return(tmp);
}
public override void Setup(IEvolutionState state, IParameter paramBase)
{
// very important, remember this
base.Setup(state, paramBase);
// I'm not using the default base for any of this stuff;
// it's not safe I think.
// set up our input
Input = (LawnmowerData)state.Parameters.GetInstanceForParameterEq(
paramBase.Push(P_DATA), null, typeof(LawnmowerData));
Input.Setup(state, paramBase.Push(P_DATA));
// load our map coordinates
MaxX = state.Parameters.GetInt(paramBase.Push(P_X), null, 1);
if (MaxX == 0)
{
state.Output.Error("The width (x dimension) of the lawn must be >0",
paramBase.Push(P_X));
}
MaxY = state.Parameters.GetInt(paramBase.Push(P_Y), null, 1);
if (MaxY == 0)
{
state.Output.Error("The length (y dimension) of the lawn must be >0",
paramBase.Push(P_Y));
}
state.Output.ExitIfErrors();
// set up the map
Map = TensorFactory.Create <int>(MaxX, MaxY); // new int[maxx][maxy];
for (var x = 0; x < MaxX; x++)
{
for (var y = 0; y < MaxY; y++)
{
Map[x][y] = UNMOWED;
}
}
}
public virtual void BeforeCoevolutionaryEvaluation(IEvolutionState state, Population pop, IGroupedProblem prob)
{
if (state.Generation == 0)
{
// create arrays for the elite individuals in the population at the previous generation.
// deep clone the elite individuals as random individuals (in the initial generation, nobody has been evaluated yet).
// deal with the elites
_eliteIndividuals = TensorFactory.Create <Individual>(state.Population.Subpops.Count, NumElite);
// copy the first individuals in each subpop (they are already randomly generated)
for (var i = 0; i < _eliteIndividuals.Length; i++)
{
if (NumElite > state.Population.Subpops[i].Individuals.Count)
{
state.Output.Fatal("Number of elite partners is greater than the size of the subpop.");
}
for (var j = 0; j < NumElite; j++)
{
_eliteIndividuals[i][j] = (Individual)(state.Population.Subpops[i].Individuals[j].Clone()); // just take the first N individuals of each subpopulation
}
}
// test for shuffled
if (NumShuffled > 0)
{
var size = state.Population.Subpops[0].Individuals.Count;
for (var i = 0; i < state.Population.Subpops.Count; i++)
{
if (state.Population.Subpops[i].Individuals.Count != size)
{
state.Output.Fatal("Shuffling was requested in MultiPopCoevolutionaryEvaluator, but the subpopulation sizes are not the same. " +
"Specifically, subpopulation 0 has size " + size + " but subpopulation " + i + " has size " + state.Population.Subpops[i].Individuals.Count);
}
}
}
}
}
/// <summary>
/// Build an NxN rotation matrix[row][column] with a given seed.
/// </summary>
public static double[] /* row */ [] /* column */ BuildRotationMatrix(double rotationSeed, int N)
{
IMersenneTwister rand = new MersenneTwisterFast(ROTATION_SEED);
double[] /* row */ [] /* column */ o = TensorFactory.Create <double>(N, N);
// make random values
for (var i = 0; i < N; i++)
{
for (var k = 0; k < N; k++)
{
o[i][k] = rand.NextGaussian();
}
}
// build random values
for (var i = 0; i < N; i++)
{
// extract o[i] -> no
var no = new double[N];
for (var k = 0; k < N; k++)
{
no[k] = o[i][k];
}
// go through o[i] and o[j], modifying no
for (var j = 0; j < i; j++)
{
var d = Dot(o[i], o[j]);
var val = ScalarMul(d, o[j]);
no = Sub(no, val);
}
o[i] = Normalize(no);
}
return(o);
}
double ComputeBestFitness()
{
// This is a dynamic programming kludge.
double[][] ttable = TensorFactory.Create <double>(k, 2);
double[][] ftable = TensorFactory.Create <double>(k + 1, 2);
for (int i = 0; i < 2; i++)
{
ftable[0][i] = _nodeScore[i];
}
// Case 1: the optimum hase depth at most k
for (int i = 0; i < k; i++)
{
for (int j = 0; j < 2; j++)
{
if (i == 0)
{
double max = (1 + _edgeScore[j][2]) * _nodeScore[2];
for (int h = 3; h < 6; h++)
{
double tmp = (1 + _edgeScore[j][h]) * _nodeScore[h];
if (tmp > max)
{
max = tmp;
}
}
ttable[i][j] = _nodeScore[j] + 2 * max;
}
else
{
double max = (1 + _edgeScore[j][0]) * ttable[i - 1][0];
for (int h = 1; h < 2; h++)
{
double tmp = (1 + _edgeScore[j][h]) * ttable[i - 1][h];
if (tmp > max)
{
max = tmp;
}
}
ttable[i][j] = _nodeScore[j] + 2 * max;
}
}
}
// Case 2: the optimum has depth k+1
for (int i = 1; i < k + 1; i++)
{
for (int j = 0; j < 2; j++)
{
double max = (1 + _edgeScore[j][0]) * ftable[i - 1][0];
for (int h = 1; h < 2; h++)
{
double tmp = (1 + _edgeScore[j][h]) * ftable[i - 1][h];
if (tmp > max)
{
max = tmp;
}
}
ftable[i][j] = _nodeScore[j] + 2 * max;
}
}
double best = _nodeScore[2];
for (int i = 3; i < 6; i++)
{
if (_nodeScore[i] > best)
{
best = _nodeScore[i];
}
}
for (int i = 0; i < k; i++)
{
for (int j = 0; j < 2; j++)
{
if (ttable[i][j] > best)
{
best = ttable[i][j];
}
}
}
for (int i = 0; i < 2; i++)
{
if (0.5 * ftable[k][i] > best)
{
best = 0.5 * ftable[k][i];
}
}
return(best);
}
///// <summary>
///// Sets all subpops in pop to the expected Lambda size. Does not fill new slots with individuals.
///// </summary>
//public virtual Population SetToLambda(Population pop, IEvolutionState state)
//{
// for (var x = 0; x < pop.Subpops.Length; x++)
// {
// var s = Lambda[x];
// // check to see if the array's not the right size
// if (pop.Subpops[x].Individuals.Length != s)
// // need to increase
// {
// var newinds = new Individual[s];
// Array.Copy(pop.Subpops[x].Individuals, 0, newinds, 0,
// s < pop.Subpops[x].Individuals.Length ? s : pop.Subpops[x].Individuals.Length);
// pop.Subpops[x].Individuals = newinds;
// }
// }
// return pop;
//}
public override Population BreedPopulation(IEvolutionState state)
{
// Complete 1/5 statistics for last population
if (ParentPopulation != null)
{
// Only go from 0 to Lambda-1, as the remaining individuals may be parents.
// A child C's parent's index I is equal to C / Mu[subpop].
for (var x = 0; x < state.Population.Subpops.Count; x++)
{
var numChildrenBetter = 0;
for (var i = 0; i < Lambda[x]; i++)
{
var parent = i / (Lambda[x] / Mu[x]); // note integer division
if (state.Population.Subpops[x].Individuals[i].Fitness.BetterThan(ParentPopulation.Subpops[x].Individuals[parent].Fitness))
{
numChildrenBetter++;
}
}
if (numChildrenBetter > Lambda[x] / 5.0) // note double division
{
Comparison[x] = C_OVER_ONE_FIFTH_BETTER;
}
else if (numChildrenBetter < Lambda[x] / 5.0) // note double division
{
Comparison[x] = C_UNDER_ONE_FIFTH_BETTER;
}
else
{
Comparison[x] = C_EXACTLY_ONE_FIFTH_BETTER;
}
}
}
// load the parent population
ParentPopulation = state.Population;
// MU COMPUTATION
// At this point we need to do load our population info
// and make sure it jibes with our mu info
// the first issue is: is the number of subpops
// equal to the number of mu's?
if (Mu.Length != state.Population.Subpops.Count) // uh oh
{
state.Output.Fatal("For some reason the number of subpops is different than was specified in the file (conflicting with Mu and Lambda storage).", null);
}
// next, load our population, make sure there are no subpops smaller than the mu's
for (var x = 0; x < state.Population.Subpops.Count; x++)
{
if (state.Population.Subpops[0].Individuals.Count < Mu[x])
{
state.Output.Error("Subpopulation " + x + " must be a multiple of the equivalent mu (that is, " + Mu[x] + ").");
}
}
state.Output.ExitIfErrors();
// sort evaluation to get the Mu best of each subpop
foreach (Subpopulation s in state.Population.Subpops)
{
s.Individuals.SortByFitnessDescending();
}
// now the subpops are sorted so that the best individuals
// appear in the lowest indexes.
Population newpop = state.Population.EmptyClone();
// create the count array
Count = new int[state.BreedThreads];
// divvy up the Lambda individuals to create
// how many threads do we really need? No more than the maximum number of individuals in any subpopulation
int numThreads = 0;
for (int x = 0; x < state.Population.Subpops.Count; x++)
{
numThreads = Math.Max(numThreads, Lambda[x]);
}
numThreads = Math.Min(numThreads, state.BreedThreads);
if (numThreads < state.BreedThreads)
{
state.Output.WarnOnce("Largest lambda size (" + numThreads + ") is smaller than number of breedthreads (" + state.BreedThreads +
"), so fewer breedthreads will be created.");
}
NewIndividuals = TensorFactory.Create <IList <Individual> >(state.Population.Subpops.Count, numThreads);
int[][] numinds = TensorFactory.Create <int>(numThreads, state.Population.Subpops.Count);
int[][] from = TensorFactory.Create <int>(numThreads, state.Population.Subpops.Count);
for (int x = 0; x < state.Population.Subpops.Count; x++)
{
for (int thread = 0; thread < numThreads; thread++)
{
NewIndividuals[x][thread].Clear();
}
int length = Lambda[x];
// we will have some extra individuals. We distribute these among the early subpopulations
int individualsPerThread = length / numThreads; // integer division
int slop = length - numThreads * individualsPerThread;
int currentFrom = 0;
for (int y = 0; y < numThreads; y++)
{
if (slop > 0)
{
numinds[y][x] = individualsPerThread + 1;
slop--;
}
else
{
numinds[y][x] = individualsPerThread;
}
if (numinds[y][x] == 0)
{
state.Output.WarnOnce("More threads exist than can be used to breed some subpopulations (first example: subpopulation " + x + ")");
}
from[y][x] = currentFrom;
currentFrom += numinds[y][x];
}
}
//for (var y = 0; y < state.BreedThreads; y++)
// for (var x = 0; x < state.Population.Subpops.Length; x++)
// {
// // figure numinds
// if (y < state.BreedThreads - 1)
// // not last one
// numinds[y][x] = Lambda[x] / state.BreedThreads;
// // in case we're slightly off in division
// else
// numinds[y][x] = Lambda[x] / state.BreedThreads + (Lambda[x] - (Lambda[x] / state.BreedThreads) * state.BreedThreads);
// // figure from
// from[y][x] = (Lambda[x] / state.BreedThreads) * y;
// }
if (numThreads == 1)
{
BreedPopChunk(newpop, state, numinds[0], from[0], 0);
}
else
{
ParallelBreeding(state, newpop, from, numinds, this);
}
// Coalesce
for (int subpop = 0; subpop < state.Population.Subpops.Count; subpop++)
{
IList <Individual> newpopinds = newpop.Subpops[subpop].Individuals;
for (int thread = 0; thread < numThreads; thread++)
{
((List <Individual>)newpopinds).AddRange(NewIndividuals[subpop][thread]);
}
}
return(PostProcess(newpop, state.Population, state));
}
public override int Produce(
int min,
int max,
int subpop,
IList <Individual> inds,
IEvolutionState state,
int thread,
IDictionary <string, object> misc)
{
int start = inds.Count;
// how many individuals should we make?
var n = TypicalIndsProduced;
if (n < min)
{
n = min;
}
if (n > max)
{
n = max;
}
// should we bother?
if (!state.Random[thread].NextBoolean(Likelihood))
{
// just load from source 0
Sources[0].Produce(n, n, subpop, inds, state, thread, misc);
return(n);
}
Parents.Clear();
// fill up parents:
for (var i = 0; i < Sources.Length; i++) // parents.length == sources.length
{
// produce one parent from each source
Sources[i].Produce(1, 1, subpop, Parents, state, thread, misc);
}
// We assume all of the species are the same species ...
var species = (VectorSpecies)((VectorIndividual)Parents[0]).Species;
// an array of the split points (width = 1)
var points = new int[((VectorIndividual)Parents[0]).GenomeLength - 1];
for (var i = 0; i < points.Length; i++)
{
points[i] = i + 1; // first split point/index = 1
}
// split all the parents into object arrays
var pieces = TensorFactory.Create <object>(Parents.Count, ((VectorIndividual)Parents[0]).GenomeLength);
// splitting...
for (int i = 0; i < Parents.Count; i++)
{
if (((VectorIndividual)Parents[i]).GenomeLength != ((VectorIndividual)Parents[0]).GenomeLength)
{
state.Output.Fatal("All vectors must be of the same length for crossover!");
}
else
{
((VectorIndividual)Parents[i]).Split(points, pieces[i]);
}
}
// crossing them over now
for (var i = 0; i < pieces[0].Length; i++)
{
if (state.Random[thread].NextBoolean(species.CrossoverProbability))
{
// shuffle
for (var j = pieces.Length - 1; j > 0; j--) // no need to shuffle first index at the end
{
// find parent to swap piece with
var parent2 = state.Random[thread].NextInt(j); // not inclusive; don't want to swap with self
// swap
var temp = pieces[j][i];
pieces[j][i] = pieces[parent2][i];
pieces[parent2][i] = temp;
}
}
}
// join them and add them to the population starting at the start location
for (int i = 0, q = start; i < Parents.Count; i++, q++)
{
((VectorIndividual)Parents[i]).Join(pieces[i]);
Parents[i].Evaluated = false;
//if (q < inds.Count) // just in case
//{
// inds[q] = Parents[i];
//}
// by Ermo. The comment code seems to be wrong. inds are empty, which means indes.size() returns 0.
// I think it should be changed to following code
// Sean -- right?
inds.Add(Parents[i]);
}
return(n);
}
public virtual void Preprocess(IEvolutionState state, int maxTreeSize)
{
state.Output.Message("Determining Tree Sizes");
MaxTreeSize = maxTreeSize;
var functionSetRepository = ((GPInitializer)state.Initializer).FunctionSetRepository;
// Put each function set into the arrays
FunctionSets = new GPFunctionSet[functionSetRepository.Count];
FunctionSetsHash = Hashtable.Synchronized(new Hashtable());
var e = functionSetRepository.Values.GetEnumerator();
var count = 0;
while (e.MoveNext())
{
var funcs = (GPFunctionSet)e.Current;
FunctionSetsHash[funcs] = count;
FunctionSets[count++] = funcs;
}
// For each function set, assign each GPNode to a unique integer
// so we can keep track of it (ick, this will be inefficient!)
FuncNodesHash = Hashtable.Synchronized(new Hashtable());
var t_nodes = Hashtable.Synchronized(new Hashtable());
count = 0;
MaxArity = 0;
for (var x = 0; x < FunctionSets.Length; x++)
{
GPNode n;
// hash all the nodes so we can remove duplicates
for (var typ = 0; typ < FunctionSets[x].Nodes.Length; typ++)
{
for (var nod = 0; nod < FunctionSets[x].Nodes[typ].Length; nod++)
{
t_nodes[n = FunctionSets[x].Nodes[typ][nod]] = n;
}
}
// rehash with Integers, yuck
e = t_nodes.Values.GetEnumerator();
GPNode tmpn;
while (e.MoveNext())
{
tmpn = (GPNode)e.Current;
if (MaxArity < tmpn.Children.Length)
{
MaxArity = tmpn.Children.Length;
}
if (!FuncNodesHash.ContainsKey(tmpn)) // don't remap the node; it'd make holes
{
FuncNodesHash[tmpn] = count++;
}
}
}
NumFuncNodes = FuncNodesHash.Count;
var initializer = (GPInitializer)state.Initializer;
var numAtomicTypes = initializer.NumAtomicTypes;
var numSetTypes = initializer.NumSetTypes;
var functionSetsLength = FunctionSets.Length;
var atomicPlusSetTypes = numAtomicTypes + numSetTypes;
var maxTreeSizePlusOne = MaxTreeSize + 1;
// set up the arrays
// NUMTREESOFTYPE
NUMTREESOFTYPE = TensorFactory.Create <BigInteger>(functionSetsLength, atomicPlusSetTypes, maxTreeSizePlusOne);
// NUMTREESROOTEDBYNODE
NUMTREESROOTEDBYNODE = TensorFactory.Create <BigInteger>(functionSetsLength, NumFuncNodes, maxTreeSizePlusOne);
// NUMCHILDPERMUTATIONS
NUMCHILDPERMUTATIONS = TensorFactory.Create <BigInteger>(functionSetsLength,
NumFuncNodes,
maxTreeSizePlusOne,
maxTreeSizePlusOne,
MaxArity);
// ROOT_D
ROOT_D = TensorFactory.CreateOpenEnded <UniformGPNodeStorage>(functionSetsLength,
atomicPlusSetTypes,
maxTreeSizePlusOne); // 4D OpenEnded
// ROOT_D_ZERO
ROOT_D_ZERO = TensorFactory.Create <bool>(functionSetsLength,
atomicPlusSetTypes,
maxTreeSizePlusOne);
// CHILD_D
CHILD_D = TensorFactory.CreateOpenEnded <double>(functionSetsLength,
NumFuncNodes,
maxTreeSizePlusOne,
maxTreeSizePlusOne); // 5D OpenEnded
var types = ((GPInitializer)(state.Initializer)).Types;
// _TrueSizesBigInt
TrueSizesBigInt = TensorFactory.Create <BigInteger>(functionSetsLength,
atomicPlusSetTypes,
maxTreeSizePlusOne);
// Go through each function set and determine numbers
// (this will take quite a while! Thankfully it's offline)
for (var x = 0; x < FunctionSets.Length; x++)
{
for (var y = 0; y < numAtomicTypes + numSetTypes; y++)
{
for (var z = 1; z <= MaxTreeSize; z++)
{
state.Output.Message("FunctionSet: " + FunctionSets[x].Name + ", Type: " + types[y].Name
+ ", Size: " + z + " num: " + (TrueSizesBigInt[x][y][z] = NumTreesOfType(initializer, x, y, z)));
}
}
}
state.Output.Message("Compiling Distributions");
TrueSizes = TensorFactory.Create <double>(functionSetsLength,
atomicPlusSetTypes,
maxTreeSizePlusOne);
// convert to doubles and organize distribution
for (var x = 0; x < FunctionSets.Length; x++)
{
for (var y = 0; y < numAtomicTypes + numSetTypes; y++)
{
for (var z = 1; z <= MaxTreeSize; z++)
{
TrueSizes[x][y][z] = (double)TrueSizesBigInt[x][y][z]; // BRS : DOES THIS TRUNCATE ANYTHING ???
}
// and if this is all zero (a possibility) we should be forgiving (hence the 'true') -- I *think*
RandomChoice.OrganizeDistribution(TrueSizes[x][y], true);
}
}
// compute our percentages
ComputePercentages();
}
public virtual void PerformCoevolutionaryEvaluation(IEvolutionState state, Population pop, IGroupedProblem prob)
{
var evaluations = 0;
_inds = new Individual[pop.Subpops.Count];
_updates = new bool[pop.Subpops.Count];
// we start by warming up the selection methods
if (NumCurrent > 0)
{
for (var i = 0; i < _selectionMethodCurrent.Length; i++)
{
_selectionMethodCurrent[i].PrepareToProduce(state, i, 0);
}
}
if (NumPrev > 0)
{
for (var i = 0; i < _selectionMethodPrev.Length; i++)
{
// do a hack here
var currentPopulation = state.Population;
state.Population = _previousPopulation;
_selectionMethodPrev[i].PrepareToProduce(state, i, 0);
state.Population = currentPopulation;
}
}
// build subpopulation array to pass in each time
var subpops = new int[state.Population.Subpops.Count];
for (var j = 0; j < subpops.Length; j++)
{
subpops[j] = j;
}
// handle shuffled always
if (NumShuffled > 0)
{
int[] /*numShuffled*/ [] /*subpop*/ [] /*shuffledIndividualIndexes*/ ordering = null;
// build shuffled orderings
ordering = TensorFactory.Create <Int32>(NumShuffled, state.Population.Subpops.Count,
state.Population.Subpops[0].Individuals.Count);
for (var c = 0; c < NumShuffled; c++)
{
for (var m = 0; m < state.Population.Subpops.Count; m++)
{
for (var i = 0; i < state.Population.Subpops[0].Individuals.Count; i++)
{
ordering[c][m][i] = i;
}
if (m != 0)
{
Shuffle(state, ordering[c][m]);
}
}
}
// for each individual
for (var i = 0; i < state.Population.Subpops[0].Individuals.Count; i++)
{
for (var k = 0; k < NumShuffled; k++)
{
for (var ind = 0; ind < _inds.Length; ind++)
{
_inds[ind] = state.Population.Subpops[ind].Individuals[ordering[k][ind][i]];
_updates[ind] = true;
}
prob.Evaluate(state, _inds, _updates, false, subpops, 0);
evaluations++;
}
}
}
// for each subpopulation
for (var j = 0; j < state.Population.Subpops.Count; j++)
{
// now do elites and randoms
if (!ShouldEvaluateSubpop(state, j, 0))
{
continue; // don't evaluate this subpopulation
}
// for each individual
for (var i = 0; i < state.Population.Subpops[j].Individuals.Count; i++)
{
var individual = state.Population.Subpops[j].Individuals[i];
// Test against all the elites
for (var k = 0; k < _eliteIndividuals[j].Length; k++)
{
for (var ind = 0; ind < _inds.Length; ind++)
{
if (ind == j)
{
_inds[ind] = individual;
_updates[ind] = true;
}
else
{
_inds[ind] = _eliteIndividuals[ind][k];
_updates[ind] = false;
}
}
prob.Evaluate(state, _inds, _updates, false, subpops, 0);
evaluations++;
}
// test against random selected individuals of the current population
for (var k = 0; k < NumCurrent; k++)
{
for (var ind = 0; ind < _inds.Length; ind++)
{
if (ind == j)
{
_inds[ind] = individual;
_updates[ind] = true;
}
else
{
_inds[ind] = ProduceCurrent(ind, state, 0);
_updates[ind] = true;
}
}
prob.Evaluate(state, _inds, _updates, false, subpops, 0);
evaluations++;
}
// Test against random selected individuals of previous population
for (int k = 0; k < NumPrev; k++)
{
for (int ind = 0; ind < _inds.Length; ind++)
{
if (ind == j)
{
_inds[ind] = individual;
_updates[ind] = true;
}
else
{
_inds[ind] = ProducePrevious(ind, state, 0);
_updates[ind] = false;
}
}
prob.Evaluate(state, _inds, _updates, false, subpops, 0);
evaluations++;
}
}
}
// now shut down the selection methods
if (NumCurrent > 0)
{
for (var i = 0; i < _selectionMethodCurrent.Length; i++)
{
_selectionMethodCurrent[i].FinishProducing(state, i, 0);
}
}
if (NumPrev > 0)
{
for (var i = 0; i < _selectionMethodPrev.Length; i++)
{
// do a hack here
var currentPopulation = state.Population;
state.Population = _previousPopulation;
_selectionMethodPrev[i].FinishProducing(state, i, 0);
state.Population = currentPopulation;
}
}
state.Output.Message("Evaluations: " + evaluations);
}
/// <summary>
/// A simple breeder that doesn't attempt to do any cross-
/// population breeding. Basically it applies pipelines,
/// one per thread, to various subchunks of a new population.
/// </summary>
public override Population BreedPopulation(IEvolutionState state)
{
Population newpop;
if (ClonePipelineAndPopulation)
{
newpop = (Population)state.Population.EmptyClone();
}
else
{
if (BackupPopulation == null)
{
BackupPopulation = (Population)state.Population.EmptyClone();
}
newpop = BackupPopulation;
newpop.Clear();
BackupPopulation = state.Population; // swap in
}
// maybe resize?
for (int i = 0; i < state.Population.Subpops.Count; i++)
{
if (ReduceBy[i] > 0)
{
int prospectiveSize = Math.Max(
Math.Max(state.Population.Subpops[i].Individuals.Count - ReduceBy[i], MinimumSize[i]),
NumElites(state, i));
if (prospectiveSize < state.Population.Subpops[i].Individuals.Count) // let's resize!
{
state.Output.Message("Subpop " + i + " reduced " + state.Population.Subpops[i].Individuals.Count + " -> " + prospectiveSize);
newpop.Subpops[i].Truncate(prospectiveSize);
}
}
}
// load Elites into top of newpop
LoadElites(state, newpop);
// how many threads do we really need? No more than the maximum number of individuals in any subpopulation
int numThreads = 0;
for (int x = 0; x < state.Population.Subpops.Count; x++)
{
numThreads = Math.Max(numThreads, state.Population.Subpops[x].Individuals.Count);
}
numThreads = Math.Min(numThreads, state.BreedThreads);
if (numThreads < state.BreedThreads)
{
state.Output.WarnOnce("Largest subpopulation size (" + numThreads + ") is smaller than number of breedthreads (" + state.BreedThreads +
"), so fewer breedthreads will be created.");
}
NewIndividuals = TensorFactory.Create <IList <Individual> >(state.Population.Subpops.Count, numThreads);
for (int subpop = 0; subpop < state.Population.Subpops.Count; subpop++)
{
for (int thread = 0; thread < numThreads; thread++)
{
NewIndividuals[subpop][thread] = new List <Individual>();
}
}
int[][] numinds = TensorFactory.Create <int>(state.BreedThreads, state.Population.Subpops.Count);
int[][] from = TensorFactory.Create <int>(state.BreedThreads, state.Population.Subpops.Count);
for (int x = 0; x < state.Population.Subpops.Count; x++)
{
for (int thread = 0; thread < numThreads; thread++)
{
NewIndividuals[x][thread].Clear();
}
int length = ComputeSubpopulationLength(state, x, 0);
// we will have some extra individuals. We distribute these among the early subpopulations
int individualsPerThread = length / numThreads; // integer division
int slop = length - numThreads * individualsPerThread;
int currentFrom = 0;
for (int y = 0; y < numThreads; y++)
{
if (slop > 0)
{
numinds[y][x] = individualsPerThread + 1;
slop--;
}
else
{
numinds[y][x] = individualsPerThread;
}
if (numinds[y][x] == 0)
{
state.Output.WarnOnce("More threads exist than can be used to breed some subpopulations (first example: subpopulation " + x + ")");
}
from[y][x] = currentFrom;
currentFrom += numinds[y][x];
}
}
if (numThreads == 1)
{
BreedPopChunk(newpop, state, numinds[0], from[0], 0);
}
else
{
ParallelBreeding(state, newpop, from, numinds, this);
}
// Coalesce
for (int subpop = 0; subpop < state.Population.Subpops.Count; subpop++)
{
IList <Individual> newpopindividuals = newpop.Subpops[subpop].Individuals;
for (int thread = 0; thread < numThreads; thread++)
{
((List <Individual>)newpopindividuals).AddRange(NewIndividuals[subpop][thread]);
}
}
return(newpop);
}
/// <summary>
/// Tests an individual, returning its successful positives in totpos and its successful negatives in totneg.
/// </summary>
/// <param name="state"></param>
/// <param name="ind"></param>
/// <param name="threadnum"></param>
/// <param name="pos"></param>
/// <param name="neg"></param>
public void FullTest(IEvolutionState state, Individual ind, int threadnum, bool[][] pos, bool[][] neg)
{
// reset the graph
NumNodes = 2;
NumEdges = 1; From[0] = 0; To[0] = 1;
Start[0] = Start[1] = Accept[0] = Accept[1] = false;
Input.edge = 0;
// generate the graph
((GPIndividual)ind).Trees[0].Child.Eval(
state, threadnum, Input, Stack, ((GPIndividual)ind), this);
// produce the adjacency matrix
if (Reading1.Length < NumNodes ||
Reading1[0].Length < NumEdges)
{
Reading1 = TensorFactory.Create <int>(NumNodes * 2, NumEdges * 2); // new int[numNodes * 2][numEdges * 2];
Reading0 = TensorFactory.Create <int>(NumNodes * 2, NumEdges * 2); // new int[numNodes*2][numEdges*2];
Epsilon = TensorFactory.Create <int>(NumNodes * 2, NumEdges * 2); // new int[numNodes*2][numEdges*2];
Reading1L = new int[NumNodes * 2];
Reading0L = new int[NumNodes * 2];
EpsilonL = new int[NumNodes * 2];
}
for (int y = 0; y < NumNodes; y++)
{
Reading1L[y] = 0;
Reading0L[y] = 0;
EpsilonL[y] = 0;
}
for (var y = 0; y < NumEdges; y++)
{
switch (Reading[y])
{
case READING0:
Reading0[From[y]][Reading0L[From[y]]++] = To[y];
break;
case READING1:
Reading1[From[y]][Reading1L[From[y]]++] = To[y];
break;
case EPSILON:
Epsilon[From[y]][EpsilonL[From[y]]++] = To[y];
break;
}
}
// create the states
if (State1.Length < NumNodes)
{
State1 = new bool[NumNodes * 2];
State2 = new bool[NumNodes * 2];
}
// test the graph on our data
_totpos = 0;
_totneg = 0;
for (var y = 0; y < pos.Length; y++)
{
if (Test(pos[y]))
{
_totpos++;
}
}
for (var y = 0; y < neg.Length; y++)
{
if (!Test(neg[y]))
{
_totneg++;
}
}
}
