More refactoring

Example/Example/Tests/Helpers/CompletionSerializer.swift

@@ -6,8 +6,6 @@
 //
 //
 
-import Foundation
-
 class CompletionSerializer {
 
     typealias CompletableFunction = (@escaping () -> ()) -> ()

Example/Example/Tests/Helpers/Float.swift

@@ -7,10 +7,20 @@
 
 import Foundation
 
-extension Array where Element == Float {
+public protocol Randomizable {
+    static func random() -> Self
+}
+
+extension Float : Randomizable {
+    public static func random() -> Float {
+        return Float(arc4random()) / Float(UInt32.max)
+    }
+}
+
+extension Array where Element : Randomizable {
 
-    static func random(count: Int) -> [Float] {
-        return (0..<count).map { _ in Float(arc4random()) / Float(UINT32_MAX) }
+    static func random(count: Int) -> [Element] {
+        return (0..<count).map { _ in Element.random() }
     }
 
 }

Example/Example/Tests/Helpers/Texture.swift

@@ -31,15 +31,15 @@ public class Texture {
             data[y * width + x][z] = newValue
         }
     }
-    
+
     var width: Int {
         return size.w
     }
-    
+
     var height: Int {
         return size.h
     }
-    
+
     var depth: Int {
         return size.f
     }
@@ -164,5 +164,5 @@ extension Texture {
         }
         return texture
     }
-    
+
 }

Example/Example/Tests/InstanceNormTests.swift

@@ -26,8 +26,8 @@ class InstanceNormTest: BenderTest {
 
     func test(texture: Texture, completion: @escaping () -> ()) {
         let styleNet = Network(inputSize: texture.size)
-        let weights = [Float].init(repeating: Float(arc4random()) / Float(UINT32_MAX), count: texture.depth)
-        let bias = [Float].init(repeating: Float(arc4random()) / Float(UINT32_MAX), count: texture.depth)
+        let weights = [Float].init(repeating: Float.random(), count: texture.depth)
+        let bias = [Float].init(repeating: Float.random(), count: texture.depth)
         let scale = Data.init(bytes: weights, count: texture.totalCount * MemoryLayout<Float>.stride)
         let shift = Data.init(bytes: bias, count: texture.totalCount * MemoryLayout<Float>.stride)
         styleNet.start ->> InstanceNorm(scale: scale, shift: shift)
@@ -100,5 +100,5 @@ class InstanceNormTest: BenderTest {
 //        }
         return output
     }
-    
+
 }

Example/Example/ViewController.swift

@@ -11,7 +11,7 @@ import UIKit
 class TestViewController: UIViewController {
 
     let testRunner = BenderTestRunner()
-    
+
     override func viewDidLoad() {
         super.viewDidLoad()
         testRunner.run()

Example/MNISTTestController.swift

@@ -55,7 +55,7 @@ class MNISTTestController: UIViewController, ExampleViewController {
 
         // If you want to create it from scratch comment the line above and uncomment:
         // createMNISTNetwork()
-        
+
         var me = self
         me.setPixelBufferPool()
         setupMetalView()
@@ -197,7 +197,7 @@ class MNISTTestController: UIViewController, ExampleViewController {
         }
 
     }
-    
+
 }
 
 // MARK: - AVCaptureVideoDataOutputSampleBufferDelegate
@@ -231,7 +231,7 @@ extension MNISTTestController: AVCaptureVideoDataOutputSampleBufferDelegate {
         } else {
             debugPrint("samplebuffer is nil \(sampleBuffer)")
         }
-        
+
         return nil
     }
 

Example/RandomLoader.swift

@@ -5,7 +5,6 @@
 //  Created by Mathias Claassen on 5/24/17.
 //
 //
-
 import Foundation
 import MetalBender
 
@@ -24,13 +23,9 @@ class RandomParameterLoader: ParameterLoader {
         free(pointer)
     }
 
-    func random() -> Float {
-        return Float(Double(arc4random()) / Double(UINT32_MAX))
-    }
-
     func uniformRandom(_ x: UnsafeMutablePointer<Float>, count: Int, scale: Float) {
         for i in 0..<count {
-            x[i] = (random()*2 - 1) * scale
+            x[i] = (Float.random()*2 - 1) * scale
         }
     }
 

Sources/Adapters/TFOptimizer.swift

@@ -6,14 +6,12 @@
 //
 //
 
-import Foundation
-
 /// Processes a grpah imported from TensorFlow applying some optimizations/simplifications
 public protocol TFOptimizer {
 
     /// Optimize a grsph imported from TensorFlow. Nodes that are to be removed should be left without adjacencies
     func optimize(graph: TFGraph)
-    
+
 }
 
 public extension TFOptimizer {
@@ -22,7 +20,7 @@ public extension TFOptimizer {
     /// This information can later be used by the 'activationNeuron' function
     func addNeuronIfThere(node: TFNode) {
         let outgoing = node.outgoingNodes()
-        if outgoing.count == 1, let next = (outgoing.first as? TFNode),
+        if let next = (outgoing.first as? TFNode),
             next.nodeDef.isTFReLuOp || next.nodeDef.isTFTanhOp || next.nodeDef.isTFSigmoidOp {
             var neuron = Tensorflow_AttrValue()
             neuron.value = Tensorflow_AttrValue.OneOf_Value.s(next.nodeDef.op.data(using: .utf8)!)

Sources/Adapters/Tensorflow/ProtoExtensions.swift

@@ -25,7 +25,7 @@ extension Tensorflow_NodeDef {
             let formatString = String(data: dataFormat, encoding: .utf8) else {
             return (Int(strides[1]), Int(strides[2]))
         }
-        
+
         let strideX = formatString == "NHWC" ? strides[2] : strides[3]
         let strideY = formatString == "NHWC" ? strides[1] : strides[2]
         return (Int(strideX), Int(strideY))

Sources/Adapters/Tensorflow/String+TFParsing.swift

@@ -6,8 +6,6 @@
 //
 //
 
-import Foundation
-
 extension Tensorflow_NodeDef {
 
     var isTFAddOp: Bool { return op == Constants.Ops.Add }

Sources/Adapters/Tensorflow/TFConvOptimizer.swift

@@ -6,8 +6,6 @@
 //
 //
 
-import Foundation
-
 /// Combines Conv2d with BiasAdd. Should be executed after Variable Processor
 class TFConvOptimizer: TFOptimizer {
 
@@ -39,5 +37,5 @@ class TFConvOptimizer: TFOptimizer {
             }
         }
     }
-    
+
 }

Sources/Adapters/Tensorflow/TFConverter+Mappers.swift

@@ -216,5 +216,5 @@ public extension TFConverter {
         }
         mappers[Constants.Ops.BatchNormGlobal] = batchnormMapper
     }
-    
+
 }

Sources/Adapters/Tensorflow/TFConverter.swift

@@ -13,7 +13,7 @@ public enum ProtoFileType {
 
     case binary
     case text
-    
+
 }
 
 /// Converts a TFNode to a NetworkLayer of Bender
@@ -105,7 +105,7 @@ open class TFConverter: Converter {
     }
 
     /// Runs the mappers through all the nodes in the `graph`.
-    /// Ops that cannot be mapped are discarded. If these ops are in the main path of the graph then the resulting graph will be disconnected. 
+    /// Ops that cannot be mapped are discarded. If these ops are in the main path of the graph then the resulting graph will be disconnected.
     ///
     /// - Parameter graph: The TFGraph to be mapped
     /// - Returns: An array of mapped oprations as NetworkLayer's

Sources/Adapters/Tensorflow/TFDeleteOptimizers.swift

@@ -6,9 +6,8 @@
 //
 //
 
-import Foundation
-
 /// Strips common nodes that are used in training but not in evaluating/testing
+
 public class TFStripTrainingOps: TFOptimizer {
 
     public var regexes: [Regex] = [TFDeleteSave().regex, TFDeleteRegularizer().regex, TFDeleteInitializer().regex]
@@ -27,7 +26,7 @@ public class TFIgnoredOpsDeleter: TFOptimizer {
 
     let ops = ["NoOp", "ExpandDims", "Cast", "Squeeze", "StopGradient", "CheckNumerics", "Assert", "Equal", "All",
                "Dequantize", "RequantizationRange", "Requantize", "PlaceholderWithDefault", "Identity"]
-    
+
     public func optimize(graph: TFGraph) {
         for node in graph.nodes {
             if ops.contains(node.nodeDef.op) {
@@ -42,7 +41,7 @@ public class TFIgnoredOpsDeleter: TFOptimizer {
 public class TFDeleteSave: TFDeleteSubgraphOptimizer {
 
     public var regex: Regex = try! Regex("save(_\\d+)?/")
-    
+
 }
 
 /// Deletes 'Initializer' subgraphs
@@ -80,5 +79,5 @@ fileprivate extension String {
         let regex = try! Regex("dropout(_\\d+)?/mul")
         return regex.test(self)
     }
-    
+
 }

Sources/Adapters/Tensorflow/TFDeleteSubgraph.swift

@@ -6,12 +6,10 @@
 //
 //
 
-import Foundation
-
 /// Deletes a specific subgraph from a TFGraph
 public protocol TFDeleteSubgraphOptimizer: TFOptimizer {
 
-    /// This Regex tells if a node is in a subgraph to be deleted or not. 
+    /// This Regex tells if a node is in a subgraph to be deleted or not.
     /// If the node's name has a match for this regex then it will be considered as belonging to a subgraph
     var regex: Regex { get set }
 
@@ -37,7 +35,7 @@ public extension TFDeleteSubgraphOptimizer {
     /// Returns an identifier for a node in this graph
     func id(for node: TFNode) -> String {
         let match = regex.match(node.nodeDef.name)
-        return (node.nodeDef.name as NSString).substring(to: match.location + match.length)
+        return String(node.nodeDef.name.prefix(match.location + match.length))
     }
 
     /// Returns if the node has incoming connections to nodes outside of the subgraph
@@ -95,5 +93,5 @@ public extension TFDeleteSubgraphOptimizer {
         // wire together
         rewire(mappings: mappings)
     }
-    
+
 }

Sources/Adapters/Tensorflow/TFDenseSubstitution.swift

@@ -6,8 +6,6 @@
 //
 //
 
-import Foundation
-
 /// Transforms a MatMul and a BiasAdd into a FullyConnected. Should be executed after Variable Processor
 /// Does not work with embedded weights. Transposing of weights must be done previously on Python side.
 public class TFDenseSubstitution: TFOptimizer {
@@ -16,7 +14,7 @@ public class TFDenseSubstitution: TFOptimizer {
           MatMul --> BiasAdd [--> Neuron]
             ^           ^
         Variable    Variable
-     
+
      Returns:
         Variable -> BiasAdd(+add-ons) <- Variable
 

Sources/Adapters/Tensorflow/TFGraph.swift

@@ -6,8 +6,6 @@
 //
 //
 
-import Foundation
-
 /// Represents a graph that is imported from a TensorFlow model
 public class TFGraph: GraphProtocol {
 
@@ -34,7 +32,7 @@ public class TFGraph: GraphProtocol {
         for node in nodes {
             // Filter TF control inputs
             let filtered = node.nodeDef.input.filter { $0.first != "^" }
-            
+
             for input in filtered {
                 if let inputNode = nodesByName[input] {
                     node.addIncomingEdge(from: inputNode)
@@ -48,5 +46,5 @@ public class TFGraph: GraphProtocol {
         var me = self
         me.sortNodes()
     }
-    
+
 }

Sources/Adapters/Tensorflow/TFInstanceNormOptimizer.swift

@@ -6,8 +6,6 @@
 //
 //
 
-import Foundation
-
 /// Creates an Instance norm from a series of nodes. Must be executed after Variable processor
 /// This implementation works with the one presented in https://github.com/lengstrom/fast-style-transfer/blob/master/src/transform.py#L49  (29/05/2017)
 /// If you implement InstanceNorm differently you migth have to create your own parser.
@@ -22,7 +20,7 @@ public class TFInstanceNormOptimizer: TFDeleteSubgraphOptimizer {
              Input   -->   InstanceNormAdd  -->  Output
                                 ^       ^
          Variable -> InstanceNormMul | Variable
-     
+
      Set_of_nodes is ([Add -> Pow, Sub] -> RealDiv)
 
      */
@@ -81,5 +79,5 @@ public class TFInstanceNormOptimizer: TFDeleteSubgraphOptimizer {
             }
         }
     }
-    
+
 }

Sources/Adapters/Tensorflow/TFNode.swift

@@ -6,8 +6,6 @@
 //
 //
 
-import Foundation
-
 /// A node of a graph imported from TensorFlow
 public class TFNode: Node {
 
@@ -24,9 +22,10 @@ public class TFNode: Node {
     public func isEqual(to other: Node) -> Bool {
         return self.nodeDef == (other as? TFNode)?.nodeDef
     }
-
 }
 
-public func ==(lhs: Tensorflow_NodeDef, rhs: Tensorflow_NodeDef) -> Bool {
-    return lhs.name == rhs.name
+extension Tensorflow_NodeDef : Equatable {
+    public static func ==(lhs: Tensorflow_NodeDef, rhs: Tensorflow_NodeDef) -> Bool {
+        return lhs.name == rhs.name
+    }
 }

Sources/Adapters/Tensorflow/TFReshapeOptimizer.swift

@@ -6,8 +6,6 @@
 //
 //
 
-import Foundation
-
 /// Removes Reshape nodes
 public class TFReshapeOptimizer: TFOptimizer {