progress

Author: jinyeom

genome.go

@@ -272,9 +272,35 @@ func (g *Genome) MutateAddConn(rate float64) {
 			}
 		}
 
-		newConn := NewConnGene(selectedNode0, selectedNode1, rand.NormFloat64()*6.0)
-		g.ConnGenes = append(g.ConnGenes, newConn)
+		if g.NodeGenes[selectedNode1].Type == "input" ||
+			g.NodeGenes[selectedNode0].Type == "output" {
+			return
+		}
+
+		if !g.pathExists(selectedNode1, selectedNode0) {
+			g.ConnGenes = append(g.ConnGenes, NewConnGene(selectedNode0,
+				selectedNode1, rand.NormFloat64()*6.0))
+		}
+
+	}
+}
+
+// pathExists returns true if there is a path from the source to the
+// destination. Helper method of MutateAddConn.
+func (g *Genome) pathExists(src, dst int) bool {
+	if src == dst {
+		return true
+	}
+
+	for _, edge := range g.ConnGenes {
+		if edge.From == src {
+			if g.pathExists(edge.To, dst) {
+				return true
+			}
+		}
 	}
+
+	return false
 }
 
 // Crossover returns a new child genome by performing crossover between the two

neat.go

@@ -184,7 +184,7 @@ func (n *NEAT) Reproduce() {
 		// reproduction of this species is only executed, if there is enough room.
 		if numSurvived > 2 && numEliminated > 0 {
 			// adjust the fitness of each member genome of this species.
-			s.ExplicitFitnessSharing()
+			//s.ExplicitFitnessSharing()
 
 			sort.Slice(s.Members, func(i, j int) bool {
 				return n.Comparison(s.Members[i], s.Members[j])
@@ -248,8 +248,8 @@ func (n *NEAT) randActivationFunc() *ActivationFunc {
 	return n.Activations[rand.Intn(len(n.Activations))]
 }
 
-// Run executes evolution.
-func (n *NEAT) Run() {
+// Run executes evolution and return the best genome.
+func (n *NEAT) Run() *Genome {
 	if n.Config.Verbose {
 		n.Config.Summarize()
 	}
@@ -258,6 +258,18 @@ func (n *NEAT) Run() {
 	for i := 0; i < n.Config.NumGenerations; i++ {
 		n.Evaluate()
 
+		// update the best genome
+		for _, genome := range n.Population {
+			if n.Comparison(genome, n.Best) {
+				n.Best = genome.Copy()
+			}
+		}
+
+		n.Statistics.Update(i, n)
+		if n.Config.Verbose {
+			n.Summarize(i)
+		}
+
 		// speciate genomes and reproduce children genomes
 		n.Speciate()
 		n.Reproduce()
@@ -273,17 +285,7 @@ func (n *NEAT) Run() {
 			}
 			n.Species = survived
 		}
-
-		// update the best genome
-		for _, genome := range n.Population {
-			if n.Comparison(genome, n.Best) {
-				n.Best = genome.Copy()
-			}
-		}
-
-		n.Statistics.Update(i, n)
-		if n.Config.Verbose {
-			n.Summarize(i)
-		}
 	}
+
+	return n.Best
 }

neat_test.go

@@ -18,9 +18,9 @@ func NEATUnitTest() {
 
 	fmt.Println("\x1b[32m=Testing NEAT with XOR test...\x1b[0m")
 	n0 := New(configXOR, XORTest())
-	n0.Run()
+	best := n0.Run()
 
-	nn := NewNeuralNetwork(n0.Best)
+	nn := NewNeuralNetwork(best)
 	output, _ := nn.FeedForward([]float64{1.0, 1.0, 1.0})
 	fmt.Println(output)
 	output, _ = nn.FeedForward([]float64{1.0, 0.0, 1.0})
@@ -30,19 +30,22 @@ func NEATUnitTest() {
 	output, _ = nn.FeedForward([]float64{1.0, 0.0, 0.0})
 	fmt.Println(output)
 
-	fmt.Println("\x1b[32m=Testing NEAT with pole balancing test...\x1b[0m")
-	configPole, err := NewConfigJSON("config_pole_balancing.json")
-	if err != nil {
-		fmt.Println("\x1b[31mFAIL\x1b[0m")
-	}
-	configPole.Summarize()
-	n1 := New(configPole, PoleBalancingTest(false, 120000))
-	n1.Run()
-
-	fmt.Println("\x1b[32m=Testing NEAT with pole balancing (random)...\x1b[0m")
-	configPole.Summarize()
-	n2 := New(configPole, PoleBalancingTest(true, 120000))
-	n2.Run()
+	/*
+		fmt.Println("\x1b[32m=Testing NEAT with pole balancing test...\x1b[0m")
+		configPole, err := NewConfigJSON("config_pole_balancing.json")
+		if err != nil {
+			fmt.Println("\x1b[31mFAIL\x1b[0m")
+		}
+		configPole.Summarize()
+		n1 := New(configPole, PoleBalancingTest(false, 120000))
+		n1.Run()
+
+		fmt.Println("\x1b[32m=Testing NEAT with pole balancing (random)...\x1b[0m")
+		configPole.Summarize()
+		n2 := New(configPole, PoleBalancingTest(true, 120000))
+		n2.Run()
+
+	*/
 }
 
 func TestNEAT(t *testing.T) {

neural_network.go

@@ -156,6 +156,7 @@ func (n *NeuralNetwork) FeedForward(inputs []float64) ([]float64, error) {
 
 	// reset all neurons
 	for _, neuron := range n.Neurons {
+		neuron.Signal = 0.0
 		neuron.activated = false
 	}