public ChordChainGeneticFunction(
DiscreteNeuralNetworkByChord dnn,
NGramGraphMarkovChain <Chord> chord,
INGramWeightAssigner <Chord> assigner,
ChordRandomFunctionType type = ChordRandomFunctionType.NoRandomSelection,
ChordCrossFunctionType cross = ChordCrossFunctionType.DiscreteChoice)
{
dnn.NullCheck();
chord.NullCheck();
assigner.NullCheck();
this.NeuralNetwork = dnn;
this.NGramGraph = chord;
this.RandomFunctionType = type;
this.CrossFunctionType = cross;
this.ChordDepth = this.NeuralNetwork.InputsCount * 2;
this.NGramDepth = 10;
this.MutateCoefficient = 0.05D;
this.Assigner = assigner;
this.random = new Random();
switch (type)
{
case ChordRandomFunctionType.AllowRandomSelection:
this.RandomSelectionCoefficient = .05D;
return;
}
}
public NGramBaseRandomGraphWalker(INGramWeightAssigner <T> assighner)
{
assighner.NullCheck();
this.WeightAssigner = assighner;
this.Depth = 3;
}
private async void startMusicGeneration_Click(object sender, EventArgs e)
{
this.model = TemperaryVariables.Graph;
if (null == this.model || null == this.neuralNetwork)
{
MessageBox.Show("Load neccessary files");
}
INGramWeightAssigner <Chord> assi = null;
if (this.useWeightAssignerCheckBox.Checked)
{
NGramLinearSizeVsTimeWeightAssigner <Chord> sizetime = new NGramLinearSizeVsTimeWeightAssigner <Chord>(1F);
NGramLinearSizeDistrubutionAssigner <Chord> size = new NGramLinearSizeDistrubutionAssigner <Chord>(this.model);
Dictionary <INGramWeightAssigner <Chord>, float> dic = new Dictionary <INGramWeightAssigner <Chord>, float>()
{
{ sizetime, this.sizeVsWeightAssigner.Value / 100F },
{ size, this.countDistAssigner.Value / 100F }
};
assi = new NGramMetaWeightAssigner <Chord>(dic);
}
else
{
assi = new NGramIDWeightAssigner <Chord>(this.model);
}
this.trainingThread = new Thread(this.GeneticSearch);
this.trainingThread.Start(assi);
}
public IReadOnlyCollection <NGram <Chord>[]> Themes(INGramWeightAssigner <Chord> assigner)
{
NGramGraphChainRetriever <Chord> mostProbable = new NGramGraphChainRetriever <Chord>(this.MarkovGraph);
var chains = mostProbable.FromEachNodeRandom(assigner, this.WalkerDepth, 3);
//KMeans<NGram<Chord>> cluster = new KMeans<NGram<Chord>>(chains.ToArray(), new LevenshteinDistance<NGram<Chord>>(), 500);
//cluster.RunFrames(2);
DiscreteNeuralNetworkByChord teacher = DiscreteNeuralNetworkByChord.Load(@"C:\Users\armen_000\Documents\Visual Studio 2013\Projects\Improvisation\Improvisation\bin\Debug\Learning\eminemNN.txt");
ChordChainGeneticFunction function = new ChordChainGeneticFunction(
teacher,
this.MarkovGraph,
assigner,
ChordChainGeneticFunction.ChordRandomFunctionType.AllowRandomSelection,
ChordChainGeneticFunction.ChordCrossFunctionType.DiscreteChoice)
{
RandomSelectionCoefficient = 0.3D
};
GeneticAlgorithm <NGram <Chord>[]> genetic = new GeneticAlgorithm <NGram <Chord>[]>(
function,
new GeneticSettings(0.1F, 0.05f, 5000, GeneticSettings.OrderOfEvolution.MutateCrossover),
chains.Take(500));
for (int i = 0; i < 3; i++)
{
try
{
var l = genetic.SingleEvolutionaryCycle();
}
catch (Exception e)
{
continue;
}
// KMeans<NGram<Chord>> chords = new KMeans<NGram<Chord>>(
// genetic.CurrentPopulation.Select(x => x.Value).RandomValues(1000),
// new GaussianNoiseDistance<NGram<Chord>>(new ChordHammingDistance(), 010F), 1000);
// chords.RunFrames(3);
// genetic.SubstitutePopulation(chords.Centers);
}
return(genetic.ToList().AsReadOnly());
}
private NGram <T>[] NodeRandom(INGramWeightAssigner <T> assigner, NGram <T> nGram, int depth)
{
List <NGram <T> > ret = new List <NGram <T> >(depth);
ret.Add(nGram);
if (depth <= 0)
{
return(ret.ToArray());
}
else
{
var assign = assigner.NextPossibleStateAssignment(nGram.AsEnumerableObject().ToArray(), this.graph);
if (!assign.Any())
{
return(ret.ToArray());
}
ret.AddRange(this.NodeRandom(assigner, assign.PickRandomFromProbabilityDistrubutionsSafe(), --depth));
return(ret.ToArray());
}
}
public List <NGram <T>[]> FromEachNodeRandom(INGramWeightAssigner <T> assigner, int depth, int perNode)
{
ConcurrentBag <NGram <T>[]> ret = new ConcurrentBag <NGram <T>[]>();
Parallel.ForEach(this.graph, (item, breaka) =>
{
if (ret.Count < this.Max)
{
for (int i = 0; i < perNode; i++)
{
ret.Add(this.NodeRandom(assigner, item.Key, depth));
}
}
else
{
breaka.Break();
}
});
return(ret.ToList());
}
public List <NGram <T>[]> FromEachNodeRandom(INGramWeightAssigner <T> assigner, int depth)
{
return(this.FromEachNodeRandom(assigner, depth, 1));
}