Merge pull request #3 from Somnibyte/neural-network-revision

Neural network revision

Author: Guled

Example/MLKit/GameScene.swift

@@ -8,6 +8,7 @@
 
 import SpriteKit
 import MachineLearningKit
+import Upsurge
 
 class GameScene: SKScene, SKPhysicsContactDelegate {
 
@@ -26,7 +27,7 @@ class GameScene: SKScene, SKPhysicsContactDelegate {
     var generationCounter = 1
 
     /// Variable to keep track of the current time (this is used to determine fitness later)
-    var currentTime = NSDate()
+    var currentTimeForFlappyBird = NSDate()
 
     /// The red square (our goal area)
     var goalArea: SKShapeNode!
@@ -43,6 +44,18 @@ class GameScene: SKScene, SKPhysicsContactDelegate {
     /// The best birds from the previous generation
     var lastBestGen: [FlappyGenome] = []
 
+    /// SKLabel
+    var generationLabel: SKLabelNode!
+    var fitnessLabel: SKLabelNode!
+
+
+    /// Best score (regardless of generation)
+    var bestScore: Int = 0
+
+    /// Label that displays the best score (bestScore attribute)
+    var bestScoreLabel: SKLabelNode!
+
+
     // END of ADDITIONS
 
     let verticalPipeGap = 150.0
@@ -173,6 +186,26 @@ class GameScene: SKScene, SKPhysicsContactDelegate {
         scoreLabelNode.text = String(score)
         self.addChild(scoreLabelNode)
 
+
+        bestScore = 0
+        bestScoreLabel = SKLabelNode(fontNamed:"MarkerFelt-Wide")
+        bestScoreLabel.position = CGPoint( x: self.frame.midX - 120.0, y: 3 * self.frame.size.height / 4 + 110.0 )
+        bestScoreLabel.zPosition = 100
+        bestScoreLabel.text = "bestScore: \(self.bestScore)"
+        self.addChild(bestScoreLabel)
+
+        generationLabel = SKLabelNode(fontNamed:"MarkerFelt-Wide")
+        generationLabel.position = CGPoint( x: self.frame.midX - 150.0, y: 3 * self.frame.size.height / 4 + 140.0 )
+        generationLabel.zPosition = 100
+        generationLabel.text = "Gen: 1"
+        self.addChild(generationLabel)
+
+        fitnessLabel = SKLabelNode(fontNamed:"MarkerFelt-Wide")
+        fitnessLabel.position = CGPoint( x: self.frame.midX + 110.0, y: 3 * self.frame.size.height / 4 + 140.0 )
+        fitnessLabel.zPosition = 100
+        fitnessLabel.text = "Fitness: 0"
+        self.addChild(fitnessLabel)
+
         // Set the current bird
         currentBird = flappyBirdGenerationContainer?[currentFlappy]
 
@@ -256,8 +289,8 @@ class GameScene: SKScene, SKPhysicsContactDelegate {
 
         // Calculate the amount of time it took until the AI lost.
         let endDate: NSDate = NSDate()
-        let timeInterval: Double = endDate.timeIntervalSince(currentTime as Date)
-        currentTime = NSDate()
+        let timeInterval: Double = endDate.timeIntervalSince(currentTimeForFlappyBird as Date)
+        currentTimeForFlappyBird = NSDate()
 
         // Evaluate the current birds fitness
         currentBird?.generateFitness(score: score, time: Float(timeInterval))
@@ -267,36 +300,39 @@ class GameScene: SKScene, SKPhysicsContactDelegate {
         print("GENERATION: \(generationCounter)")
         print("BIRD COUNT: \(currentFlappy)")
         print("FITNESS: \(currentBird?.fitness)")
+        self.generationLabel.text = "Gen: \(self.generationCounter)"
         print("--------------------------- \n")
 
-        /* DEBUGGING
-        if (currentBird?.fitness)! >= Float(7.0) {
-            print("-----BEST BIRD FOUND------")
-
-            currentBird?.brain?.inputLayer.printLayer(layer: (currentBird?.brain?.inputLayer)!)
-            currentBird?.brain?.hiddenLayer.printLayer(listOfHiddenLayers: (currentBird?.brain?.listOfHiddenLayers)!)
-            currentBird?.brain?.outputLayer.printLayer(layer: (currentBird?.brain?.outputLayer)!)
-
-        }
-        */
-
         // Go to next flappy bird
         currentFlappy += 1
 
         // Filter out the worst birds after gen 6 (can be adjusted)
+
+        if let bird = currentBird {
+            if bird.fitness >= 9.0 {
+                print("FOUND RARE BIRD")
+                print(bird.brain?.layers[0].weights)
+                print(bird.brain?.layers[1].weights)
+                print(bird.brain?.layers[0].bias)
+                print(bird.brain?.layers[1].bias)
+            }
+        }
+
         if generationCounter >= 3 {
 
             // Experiment: Keep some of the last best birds and put them back into the population
-            lastBestGen = (flappyBirdGenerationContainer?.filter({$0.fitness >= 4}))!
+            lastBestGen = (flappyBirdGenerationContainer?.filter({$0.fitness >= 7.0}))!
         }
 
-        if (currentBird?.fitness)! > maxFitness {
-            maxFitness = (currentBird?.fitness)!
-            maxBird = currentBird
+        if let bird = currentBird {
+            if bird.fitness > maxFitness {
+                maxFitness = bird.fitness
+                maxBird = bird
+            }
         }
 
         // If we have hit the 20th bird, we need to move on to the next generation
-        if currentFlappy == 10 {
+        if currentFlappy == 20 {
 
             print("GENERATING NEW GEN!")
 
@@ -308,7 +344,7 @@ class GameScene: SKScene, SKPhysicsContactDelegate {
             newGen += lastBestGen
             newGen.append(maxBird!)
 
-            while newGen.count < 10 {
+            while newGen.count < 20 {
 
                 // Select the best parents
                 let parents = PopulationManager.selectParents(genomes: flappyBirdGenerationContainer!)
@@ -320,10 +356,10 @@ class GameScene: SKScene, SKPhysicsContactDelegate {
                 var offspring = BiologicalProcessManager.onePointCrossover(crossoverRate: 0.5, parentOneGenotype: parents.0.genotypeRepresentation, parentTwoGenotype: parents.1.genotypeRepresentation)
 
                 // Mutate their genes
-
                 BiologicalProcessManager.inverseMutation(mutationRate: 0.7, genotype: &offspring.0)
                 BiologicalProcessManager.inverseMutation(mutationRate: 0.7, genotype: &offspring.1)
 
+
                 // Create a separate neural network for the birds based on their genes
                 let brainofOffspring1 = GeneticOperations.decode(genotype: offspring.0)
 
@@ -346,12 +382,19 @@ class GameScene: SKScene, SKPhysicsContactDelegate {
             flappyBirdGenerationContainer = newGen
 
             // Set the current bird
-            currentBird = flappyBirdGenerationContainer?[currentFlappy]
+            if (flappyBirdGenerationContainer?.count)! > currentFlappy {
+                currentBird = flappyBirdGenerationContainer?[currentFlappy]
+            } else {
+                currentBird = maxBird
+            }
 
         } else {
 
             // Set the current bird
-            currentBird = flappyBirdGenerationContainer?[currentFlappy]
+            if (flappyBirdGenerationContainer?.count)! > currentFlappy {
+                currentBird = flappyBirdGenerationContainer?[currentFlappy]
+            }
+
         }
 
     }
@@ -385,6 +428,10 @@ class GameScene: SKScene, SKPhysicsContactDelegate {
 
     override func update(_ currentTime: TimeInterval) {
 
+        let endDate: NSDate = NSDate()
+        let timeInterval: Double = endDate.timeIntervalSince(currentTimeForFlappyBird as Date)
+        self.fitnessLabel.text = "Fitness: \(NSString(format: "%.2f", timeInterval))"
+
         checkIfOutOfBounds(bird.position.y)
 
         /* Called before each frame is rendered */
@@ -418,21 +465,14 @@ class GameScene: SKScene, SKPhysicsContactDelegate {
 
             let normalizedPosToGap = (posToGap - (-439))/(279 - (-439))
 
-            /*
-            print("======================== \n")
-            print(" DIST: \((finalDistanceOfNextPipe))")
-            print("PIPE POSITION: \(finalPosToGap)")
-            print("Bird POS Y: \(birdPos)")
-            print("======================== \n")
-            */
-
             // Decision AI makes
-            let decision = (currentBird?.brain?.forward(input: [Float(1), Float(normalizedDistanceOfNextPipe), Float(normalizedPosToGap), Float(birdYPos), Float(normalizedDistanceFromBottomPipe)]))!
+            let input = Matrix<Float>(rows: 4, columns: 1, elements: [Float(normalizedDistanceOfNextPipe), Float(normalizedPosToGap), Float(birdYPos), Float(normalizedDistanceFromBottomPipe)])
+            let decision = currentBird?.brain?.feedforward(input: input)
 
             print("FLAPPY BIRD DECISION: \(decision)")
 
             // 0.95 was arbitrary, tweaking is recommended
-            if  decision[0] >= Float(0.89) {
+            if  (decision?.elements[0])! >= Float(0.5) {
 
                 if moving.speed > 0 {
 
@@ -491,6 +531,12 @@ class GameScene: SKScene, SKPhysicsContactDelegate {
                 score += 1
                 scoreLabelNode.text = String(score)
 
+                // Update best score label
+                if score > bestScore {
+                    bestScore = score
+                    bestScoreLabel.text = "Best Score: \(self.bestScore)"
+                }
+
                 // Add a little visual feedback for the score increment
                 scoreLabelNode.run(SKAction.sequence([SKAction.scale(to: 1.5, duration:TimeInterval(0.1)), SKAction.scale(to: 1.0, duration:TimeInterval(0.1))]))
             } else {

Example/MLKit/GameViewController.swift

@@ -41,9 +41,9 @@ public class FlappyGenome: Genome {
 
     public var fitness: Float = 0
 
-    public var brain: NeuralNet?
+    public var brain: NeuralNetwork?
 
-    public init(genotype: [Float], network: NeuralNet) {
+    public init(genotype: [Float], network: NeuralNetwork) {
 
         self.genotypeRepresentation = genotype
         self.brain = network
@@ -64,14 +64,12 @@ class GameViewController: UIViewController {
         // Create First Generation of Flappy Birds
         var generation1: [FlappyGenome] = []
 
-        for _ in 1...10 {
+        for _ in 1...20 {
 
             // Bias already included
-            let brain = NeuralNet(numberOfInputNeurons: 4, hiddenLayers: [4], numberOfOutputNeurons: 1)
-
-            brain.activationFuncType = .siglog
-
-            brain.activationFuncTypeOfOutputLayer = .siglog
+            let brain = NeuralNetwork(size: (4, 1))
+            brain.addLayer(layer: Layer(size: (4, 4), activationType: .siglog))
+            brain.addLayer(layer: Layer(size: (4, 1), activationType: .siglog))
 
             let newBird = FlappyGenome(genotype: GeneticOperations.encode(network: brain), network: brain)
 

Example/MLKit/GeneticOperations.swift

@@ -11,40 +11,29 @@ import MachineLearningKit
 import Upsurge
 
 /// The GeneticOperations class manages encoding genes into weights for the neural network and decoding neural network weights into genes. These methods are not provided in the framework itself, rather it was for the game example.
-final class GeneticOperations {
 
+final class GeneticOperations {
     /**
-     The encode method converts a NueralNet object to an array of floats by taking the weights of each layer and placing them into an array.
+     The encode method converts a NueralNetwork object to an array of floats by taking the weights of each layer and placing them into an array.
 
      - parameter network: A NeuralNet Object.
 
      - returns: An array of Float values.
      */
-    public static func encode(network: NeuralNet) -> [Float] {
-
-        let inputLayerNeurons = network.inputLayer.listOfNeurons
-        let outputLayerNeurons = network.outputLayer.listOfNeurons
+    public static func encode(network: NeuralNetwork) -> [Float] {
 
         var genotypeRepresentation: [Float] = []
 
-        for neuron in inputLayerNeurons {
-            genotypeRepresentation += neuron.weightsComingIn
-        }
-
-        for hiddenLayer in network.listOfHiddenLayers {
-            for neuron in hiddenLayer.listOfNeurons {
-                genotypeRepresentation += neuron.weightsComingIn
-            }
+        for layer in network.layers {
 
-            for neuron in hiddenLayer.listOfNeurons {
-                genotypeRepresentation += neuron.weightsGoingOut
-            }
+            genotypeRepresentation += Array<Float>(layer.weights!.elements)
         }
 
-        for neuron in outputLayerNeurons {
-            genotypeRepresentation += neuron.weightsGoingOut
+        for layer in network.layers {
+            genotypeRepresentation += Array<Float>(layer.bias!.elements)
         }
 
+
         return genotypeRepresentation
     }
 
@@ -55,57 +44,17 @@ final class GeneticOperations {
 
      - returns: An array of Float values.
      */
-    public static func decode(genotype: [Float]) -> NeuralNet {
+    public static func decode(genotype: [Float]) -> NeuralNetwork {
 
         // Create a new NueralNet
-        let brain = NeuralNet(numberOfInputNeurons: 4, hiddenLayers: [4], numberOfOutputNeurons: 1)
-
-        brain.activationFuncType = .siglog
-
-        brain.activationFuncTypeOfOutputLayer = .siglog
-
-        // Convert genotype back to weights for each layer
-        let inputLayerWeights: [Float] = Array<Float>(genotype[0...4])
-
-        // First is bias neuron
-        let hiddenLayerWeightsComingInForNeuron1: [Float] = Array<Float>(genotype[5...8])
-        let hiddenLayerWeightsComingInForNeuron2: [Float] = Array<Float>(genotype[9...12])
-        let hiddenLayerWeightsComingInForNeuron3: [Float] = Array<Float>(genotype[13...16])
-        let hiddenLayerWeightsComingInForNeuron4: [Float] = Array<Float>(genotype[17...20])
-        let hiddenLayerWeightsComingInForNeuron5: [Float] = Array<Float>(genotype[21...24])
-        let hiddenLayerWeightsGoingOut: [Float] = Array<Float>(genotype[25...29])
-        let outputLayerWeightGoingOut: Float = genotype[30]
-
-        for (i, neuron) in brain.inputLayer.listOfNeurons.enumerated() {
-
-            neuron.weightsComingIn = ValueArray<Float>([inputLayerWeights[i]])
-        }
-
-        for hiddenLayer in brain.listOfHiddenLayers {
-            for (i, neuron) in hiddenLayer.listOfNeurons.enumerated() {
-
-                if i == 0 {
-                    neuron.weightsComingIn = ValueArray<Float>(hiddenLayerWeightsComingInForNeuron1)
-                } else if i == 1 {
-                    neuron.weightsComingIn = ValueArray<Float>(hiddenLayerWeightsComingInForNeuron2)
-                } else if i == 2 {
-                    neuron.weightsComingIn = ValueArray<Float>(hiddenLayerWeightsComingInForNeuron3)
-                } else if i == 3 {
-                    neuron.weightsComingIn = ValueArray<Float>(hiddenLayerWeightsComingInForNeuron4)
-                } else if i == 4 {
-                    neuron.weightsComingIn = ValueArray<Float>(hiddenLayerWeightsComingInForNeuron5)
-                }
-            }
-
-            for (i, neuron) in hiddenLayer.listOfNeurons.enumerated() {
-
-                neuron.weightsGoingOut = ValueArray<Float>([hiddenLayerWeightsGoingOut[i]])
-
-            }
-
-        }
-
-        brain.outputLayer.listOfNeurons[0].weightsGoingOut = ValueArray<Float>([outputLayerWeightGoingOut])
+        let brain = NeuralNetwork(size: (4, 1))
+        brain.addLayer(layer: Layer(size: (4, 4), activationType: .siglog))
+        brain.addLayer(layer: Layer(size: (4, 1), activationType: .siglog))
+
+        brain.layers[0].weights = Matrix<Float>(rows: 4, columns: 4, elements: ValueArray<Float>(Array<Float>(genotype[0...15])))
+        brain.layers[0].bias = Matrix<Float>(rows: 4, columns: 1, elements: ValueArray<Float>(Array<Float>(genotype[20...23])))
+        brain.layers[1].weights = Matrix<Float>(rows: 1, columns: 4, elements: ValueArray<Float>(Array<Float>(genotype[16...19])))
+        brain.layers[1].bias = Matrix<Float>(rows: 1, columns: 1, elements: ValueArray<Float>([genotype[24]]))
 
         return brain
     }