Genetic Algorithm complete. Flappy Bird AI still in testing, but is functional. Flappy bird was able to get 2 points in 30 minutes with 30 generations

Author: Guled

Example/MLKit/GameScene.swift

@@ -9,20 +9,21 @@
 import UIKit
 import SpriteKit
 import MachineLearningKit
+import Upsurge
 
 extension SKNode {
-    class func unarchiveFromFile(_ file : String) -> SKNode? {
-        
+    class func unarchiveFromFile(_ file: String) -> SKNode? {
+
         let path = Bundle.main.path(forResource: file, ofType: "sks")
-        
+
         let sceneData: Data?
         do {
             sceneData = try Data(contentsOf: URL(fileURLWithPath: path!), options: .mappedIfSafe)
         } catch _ {
             sceneData = nil
         }
         let archiver = NSKeyedUnarchiver(forReadingWith: sceneData!)
-        
+
         archiver.setClass(self.classForKeyedUnarchiver(), forClassName: "SKScene")
         let scene = archiver.decodeObject(forKey: NSKeyedArchiveRootObjectKey) as! GameScene
         archiver.finishDecoding()
@@ -31,26 +32,27 @@ extension SKNode {
 }
 
 
-public struct FlappyGenome: GenomeProtocol {
+// ADDITIONS
+
+// Genome that represents a Flappy Bird
+public class FlappyGenome: Genome {
 
-    public var network: NeuralNet?
+    public var brain: NeuralNet?
 
-    public init(genotype: [Float], network: NeuralNet){
+    public init(genotype: [Float], network: NeuralNet) {
+
+        super.init(genotype: genotype)
 
         self.genotypeRepresentation = genotype
-        self.network = network
+        self.brain = network
     }
 
-    /// Genotype representation of the genome.
-    public var genotypeRepresentation: [Float]
-
-    /// Fitness of a particular genome.
-    public var fitness: Float?
-    
+    public func generateFitness(score: Int, time: Float) {
+        self.fitness = Float(Float(score) + time)
+    }
 }
 
-
-
+// END of ADDITIONS
 
 class GameViewController: UIViewController {
 
@@ -60,14 +62,13 @@ class GameViewController: UIViewController {
 
 
         // Create First Generation of Flappy Birds
+        var generation1: [FlappyGenome] = []
 
-        var generation1:[FlappyGenome] = []
-
-        for _ in 1...10 {
+        for _ in 1...20 {
 
             let brain = NeuralNet()
 
-            brain.initializeNet(numberOfInputNeurons: 4, numberOfHiddenLayers: 1, numberOfNeuronsInHiddenLayer: 4, numberOfOutputNeurons: 1)
+            brain.initializeNet(numberOfInputNeurons: 3, numberOfHiddenLayers: 1, numberOfNeuronsInHiddenLayer: 4, numberOfOutputNeurons: 1)
 
             brain.activationFuncType = .SIGLOG
 
@@ -81,29 +82,30 @@ class GameViewController: UIViewController {
 
         if let scene = GameScene.unarchiveFromFile("GameScene") as? GameScene {
 
-            // Set the first generation of Flappy Birds 
-            scene.firstGenerationOfFlappyBirds = generation1
+            // Set the first generation of Flappy Birds
+            scene.flappyBirdGenerationContainer = generation1
 
             // Configure the view.
             let skView = self.view as! SKView
-            skView.showsFPS = true
-            skView.showsNodeCount = true
+            skView.showsFPS = false
+            skView.showsPhysics = true
+            skView.showsNodeCount = false
 
             /* Sprite Kit applies additional optimizations to improve rendering performance */
             skView.ignoresSiblingOrder = true
-            
+
             /* Set the scale mode to scale to fit the window */
             scene.scaleMode = .aspectFill
-            
+
             skView.presentScene(scene)
         }
     }
 
-    override var shouldAutorotate : Bool {
+    override var shouldAutorotate: Bool {
         return true
     }
 
-    override var supportedInterfaceOrientations : UIInterfaceOrientationMask {
+    override var supportedInterfaceOrientations: UIInterfaceOrientationMask {
         if UIDevice.current.userInterfaceIdiom == .phone {
             return UIInterfaceOrientationMask.allButUpsideDown
         } else {
@@ -116,5 +118,5 @@ class GameViewController: UIViewController {
         // Release any cached data, images, etc that aren't in use.
     }
 
-        
+
 }

Example/MLKit/GameViewController.swift

@@ -8,12 +8,24 @@
 
 import Foundation
 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 {
 
+
+    /**
+     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.
+
+     - parameter network: A NeuralNet Object.
+
+     - returns: An array of Float values.
+     */
     public static func encode(network: NeuralNet) -> [Float] {
+
         let inputLayerNuerons = network.inputLayer.listOfNeurons
-        let hiddenLayerNeurons = network.hiddenLayer.listOfNeurons
+        let hiddenLayerNeurons = network.listOfHiddenLayers[0].listOfNeurons
         let outputLayerNeurons = network.outputLayer.listOfNeurons
 
         var genotypeRepresentation: [Float] = []
@@ -23,7 +35,11 @@ final class GeneticOperations {
         }
 
         for neuron in hiddenLayerNeurons {
-            genotypeRepresentation = genotypeRepresentation + neuron.weightsComingIn + neuron.weightsGoingOut
+            genotypeRepresentation = genotypeRepresentation + neuron.weightsComingIn
+        }
+
+        for neuron in hiddenLayerNeurons {
+            genotypeRepresentation = genotypeRepresentation + neuron.weightsGoingOut
         }
 
         for neuron in outputLayerNeurons {
@@ -33,4 +49,64 @@ final class GeneticOperations {
         return genotypeRepresentation
     }
 
+
+    /**
+     The decode method converts a genotype back to a NeuralNet object by taking each value from the genotype and mapping them to a neuron in a particular layer.
+
+     - parameter network: A NeuralNet Object.
+
+     - returns: An array of Float values.
+     */
+    public static func decode(genotype: [Float]) -> NeuralNet {
+
+        // Create a new NueralNet
+        let brain = NeuralNet()
+
+        brain.initializeNet(numberOfInputNeurons: 3, numberOfHiddenLayers: 1, numberOfNeuronsInHiddenLayer: 4, numberOfOutputNeurons: 1)
+
+        brain.activationFuncType = .SIGLOG
+
+        brain.activationFuncTypeOutputLayer = .SIGLOG
+
+        // Convert genotype back to weights for each layer
+        let inputLayerWeights: [Float] = Array<Float>(genotype[0...3])
+
+        // First is bias neuron
+        let hiddenLayerWeightsComingInForNueron1: [Float] = Array<Float>(genotype[4...7])
+        let hiddenLayerWeightsComingInForNueron2: [Float] = Array<Float>(genotype[8...11])
+        let hiddenLayerWeightsComingInForNueron3: [Float] = Array<Float>(genotype[12...15])
+        let hiddenLayerWeightsComingInForNueron4: [Float] = Array<Float>(genotype[16...19])
+        let hiddenLayerWeightsGoingOut: [Float] = Array<Float>(genotype[20...24])
+        let outputLayerWeightGoingOut: Float = genotype[25]
+
+        for (var i, var neuron) in brain.inputLayer.listOfNeurons.enumerated() {
+
+            neuron.weightsComingIn = ValueArray<Float>([inputLayerWeights[i]])
+        }
+
+        for (var i, var neuron) in brain.listOfHiddenLayers[0].listOfNeurons.enumerated() {
+
+            if i == 0 {
+                continue
+            } else if i == 1 {
+                neuron.weightsComingIn = ValueArray<Float>(hiddenLayerWeightsComingInForNueron1)
+            } else if i == 2 {
+                neuron.weightsComingIn = ValueArray<Float>(hiddenLayerWeightsComingInForNueron2)
+            } else if i == 3 {
+                neuron.weightsComingIn = ValueArray<Float>(hiddenLayerWeightsComingInForNueron3)
+            } else if i == 4 {
+                neuron.weightsComingIn = ValueArray<Float>(hiddenLayerWeightsComingInForNueron4)
+            }
+        }
+
+        for (var i, var neuron) in brain.listOfHiddenLayers[0].listOfNeurons.enumerated() {
+
+            neuron.weightsGoingOut = ValueArray<Float>([hiddenLayerWeightsGoingOut[i]])
+
+        }
+
+        brain.outputLayer.listOfNeurons[0].weightsGoingOut = ValueArray<Float>([outputLayerWeightGoingOut])
+
+        return brain
+    }
 }

Example/MLKit/GeneticOperations.swift

@@ -15,17 +15,15 @@ import Nimble
 
 class GeneticSpec: QuickSpec {
 
-
-
     override func spec() {
 
         it("Should be able to produce a unique genotype after swap mutation process.") {
 
-            var fakeGenome:Genome = Genome(genotype: [1.0,2.0,3.0])
+            let fakeGenome: Genome = Genome(genotype: [1.0, 2.0, 3.0])
 
             let oldGenotype = fakeGenome.genotypeRepresentation
 
-            BiologicalProcessManager.swapMutation(genotype: &fakeGenome.genotypeRepresentation)
+            BiologicalProcessManager.swapMutation(mutationRate: 1.0, genotype: &fakeGenome.genotypeRepresentation )
 
 
             expect(oldGenotype).toNot(equal(fakeGenome.genotypeRepresentation))
@@ -34,39 +32,40 @@ class GeneticSpec: QuickSpec {
 
         it("Should be able to produce a unique genotype after insert mutation process. ") {
 
-            var fakeGenome:Genome = Genome(genotype: [1.0,2.0,3.0,4.0,5.0,6.0,7.0])
+            let fakeGenome: Genome = Genome(genotype: [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0])
 
             let oldGenotype = fakeGenome.genotypeRepresentation
 
-            BiologicalProcessManager.insertMutation(genotype: &fakeGenome.genotypeRepresentation)
+            BiologicalProcessManager.insertMutation(mutationRate: 1.0, genotype: &fakeGenome.genotypeRepresentation )
 
             expect(oldGenotype).toNot(equal(fakeGenome.genotypeRepresentation))
         }
 
 
         it("Should be able to produce a unique genotype after inverse mutation process. ") {
 
-            var fakeGenome:Genome = Genome(genotype: [1.0,2.0,3.0])
+            let fakeGenome: Genome = Genome(genotype: [1.0, 2.0, 3.0])
 
             let oldGenotype = fakeGenome.genotypeRepresentation
 
-            BiologicalProcessManager.inverseMutation(genotype: &fakeGenome.genotypeRepresentation)
+            BiologicalProcessManager.inverseMutation(mutationRate: 1.0, genotype: &fakeGenome.genotypeRepresentation )
 
             expect(oldGenotype).toNot(equal(fakeGenome.genotypeRepresentation))
         }
 
         it("Should be able to produce a unique genotype after scramble mutation process. ") {
 
-            var fakeGenome:Genome = Genome(genotype: [1.0,2.0,3.0,4.0,5.0,6.0,7.0])
+            let fakeGenome: Genome = Genome(genotype: [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0])
 
             let oldGenotype = fakeGenome.genotypeRepresentation
 
-            BiologicalProcessManager.scrambleMutation(genotype: &fakeGenome.genotypeRepresentation)
+            BiologicalProcessManager.scrambleMutation(mutationRate: 1.0, genotype: &fakeGenome.genotypeRepresentation)
 
             expect(oldGenotype).toNot(equal(fakeGenome.genotypeRepresentation))
         }
 
 
+
     }
-    
+
 }

Example/Tests/GeneticSpec.swift

@@ -3,45 +3,8 @@
 import UIKit
 import MachineLearningKit
 
-// Slicing 
-struct test {
-    var testArray:[Int] = [1,2,3,4,5,6,7,8]
-}
+// Slicing
 
-var t = test()
+var testArray: [Int] = [1, 2,3, 4,5, 6,7, 8]
 
-func swapMutation( genotype:inout [Int]) {
-    genotype[0] = genotype[1]
-
-}
-
-
-extension Collection {
-    /// Return a copy of `self` with its elements shuffled
-    func shuffle() -> [Iterator.Element] {
-        var list = Array(self)
-        list.shuffle()
-        return list
-    }
-}
-
-extension MutableCollection where Indices.Iterator.Element == Index {
-    /// Shuffles the contents of this collection.
-    mutating func shuffle() {
-        let c = count
-        guard c > 1 else { return }
-
-        for (firstUnshuffled, unshuffledCount) in zip(indices, stride(from: c, to: 1, by: -1)) {
-            let d: IndexDistance = numericCast(arc4random_uniform(numericCast(unshuffledCount)))
-            guard d != 0 else { continue }
-            let i = index(firstUnshuffled, offsetBy: d)
-            swap(&self[firstUnshuffled], &self[i])
-        }
-    }
-}
-
-
-//t.testArray[0...3].shuffle()
-print(t.testArray)
-print(t.testArray[0...3].reverse())
-print(t.testArray)
+var t = testArray[0...3]

MLKit-PlayGround.playground/Contents.swift

@@ -12,7 +12,7 @@
 import Foundation
 import Upsurge
 
-
+/// The HiddenLayer class represents the hidden layer of a NueralNet object.
 public class HiddenLayer: Layer {
 
 

MLKit/Classes/ANN/HiddenLayer.swift

@@ -10,6 +10,7 @@
 import Foundation
 import Upsurge
 
+/// The InputLayer class represents the input layer of a NueralNet object.
 public class InputLayer: Layer, InputandOutputLayerMethods {
 
 
@@ -29,7 +30,7 @@ public class InputLayer: Layer, InputandOutputLayerMethods {
     }
 
     /// Number of neurons in the input layer
-    public var numberOfNueronsInLayer: Int { 
+    public var numberOfNueronsInLayer: Int {
 
         get {
             return _numberOfNueronsInLayer

MLKit/Classes/ANN/InputLayer.swift

@@ -12,6 +12,7 @@ import Foundation
 import Upsurge
 
 
+/// The Layer Protocol defines what attributes and methods a Layer Object must have.
 public protocol Layer {
 
     /// List of Neuron Object in a particular layer.
@@ -22,7 +23,6 @@ public protocol Layer {
 
 }
 
-
 public protocol InputandOutputLayerMethods {
 
     /**