Modified Raw -> _Raw

Modified `Raw` -> `_Raw`

Author: param087

.travis.yml

@@ -1,6 +1,6 @@
 env:
   global:
-  - SWIFT_BRANCH=tensorflow-0.4
+  - SWIFT_BRANCH=tensorflow-0.6
   # PACKAGE_VERSION will be used by travis to create the Github release tag
   - PACKAGE_VERSION=0.0.1
 
@@ -13,9 +13,9 @@ jobs:
     language: swift
     sudo: required
     install:
-    - wget https://storage.googleapis.com/swift-tensorflow-artifacts/releases/v0.4/rc2/swift-tensorflow-RELEASE-0.4-osx.pkg
-    - sudo installer -pkg swift-tensorflow-RELEASE-0.4-osx.pkg -target /
-    - export PATH="/Library/Developer/Toolchains/swift-tensorflow-RELEASE-0.4.xctoolchain/usr/bin:$PATH"
+    - wget https://storage.googleapis.com/swift-tensorflow-artifacts/releases/v0.6/rc1/swift-tensorflow-RELEASE-0.6-osx.pkg
+    - sudo installer -pkg swift-tensorflow-RELEASE-0.6-osx.pkg -target /
+    - export PATH="/Library/Developer/Toolchains/swift-tensorflow-RELEASE-0.6.xctoolchain/usr/bin:$PATH"
     - gem install jazzy
     - jazzy --min-acl internal --no-hide-documentation-coverage --theme fullwidth --output ./docs
     - rm -rf ./build

README.md

@@ -23,7 +23,7 @@ Swift library for Machine Learning based on [Swift for TensorFlow](https://githu
 Add the library to your projects dependencies in the Package.swift file as shown below.
 ```swift 
 dependencies: [
-        .package(url: "https://github.com/param087/swiftML", .exact("0.0.2")),
+        .package(url: "https://github.com/param087/swiftML", .exact("0.0.4")),
     ],
 ```
 
@@ -35,7 +35,7 @@ can import them:
 ```swift
 
 // Install the swiftML package from GitHub.
-%install '.package(url: "https://github.com/param087/swiftML", from: "0.0.2")' swiftML
+%install '.package(url: "https://github.com/param087/swiftML", from: "0.0.4")' swiftML
 
 // Install the swiftML package that's in the local directory.
 %install '.package(path: "$cwd/swiftML")' swiftML

Sources/swiftML/BernoulliNB.swift

@@ -45,7 +45,7 @@ public class BernoulliNB {
         precondition(labels.shape[0] > 0, "Labels must have a positive sample count.")
 
         // Find unique classes in target values.
-        (self.classes, self.indices) = Raw.unique(labels.flattened())
+        (self.classes, self.indices) = _Raw.unique(labels.flattened())
 
         // Initialize the classLogPrior and featureLogProb based on feature count and sample count.
         var separated = [[Tensor<Float>]]()

Sources/swiftML/GaussianNB.swift

@@ -32,7 +32,7 @@ public class GaussianNB {
         precondition(labels.shape[0] > 0, "Labels must have a positive sample count.")
 
         // Find unique classes in target values.
-        (self.classes, self.indices) = Raw.unique(labels.flattened())
+        (self.classes, self.indices) = _Raw.unique(labels.flattened())
 
         var separated = [[Tensor<Float>]]()
         self.model = Tensor<Float>(zeros: [self.classes.shape[0], data.shape[1], 2])

Sources/swiftML/Helpers.swift

@@ -6,9 +6,9 @@ extension Tensor where Scalar: TensorFlowFloatingPoint {
     /// Reference: ["Moore–Penrose inverse"](
     /// https://en.wikipedia.org/wiki/Moore%E2%80%93Penrose_inverse)
     public var pseudoinverse: Tensor {
-        let svd = Raw.svd(self)
-        let diag = Raw.diag(diagonal: svd.s)
-        return matmul(matmul(svd.v, Raw.matrixInverse(diag)), svd.u.transposed())
+        let svd = _Raw.svd(self)
+        let diag = _Raw.diag(diagonal: svd.s)
+        return matmul(matmul(svd.v, _Raw.matrixInverse(diag)), svd.u.transposed())
     }
 }
 
@@ -28,7 +28,7 @@ public func deterministicSvd<T: FloatingPoint & TensorFlowScalar>(
     _ input: Tensor<T>,
     columnBasedSignFlipping: Bool = true
 ) -> (s: Tensor<T>, u: Tensor<T>, v: Tensor<T>) {
-    var (s, u, v) = Raw.svd(input)
+    var (s, u, v) = _Raw.svd(input)
     v = v.transposed()
 
     let signs: Tensor<T>
@@ -39,14 +39,14 @@ public func deterministicSvd<T: FloatingPoint & TensorFlowScalar>(
         for i in 0..<u.shape[1] {
             colValueForSign[i] = u[Int(maxAbsCols[i].scalarized()), i]
         }
-        signs = Raw.sign(colValueForSign)
+        signs = _Raw.sign(colValueForSign)
     } else {
         let maxAbsRows = abs(v).argmax(squeezingAxis: 1)
         var rowValueForSign = Tensor<T>(zeros: [u.shape[0]])
         for i in 0..<u.shape[0] {
             rowValueForSign[i] = v[i, Int(maxAbsRows[i].scalarized())]
         }
-        signs = Raw.sign(rowValueForSign)
+        signs = _Raw.sign(rowValueForSign)
     }
     return (s, u * signs, v * signs.reshaped(to: [u.shape[1], 1]))
 }

Sources/swiftML/KMeans.swift

@@ -91,13 +91,13 @@ public class KMeans {
             }
 
             let probability = distance / distance.sum()
-            let cumsum = Raw.cumsum(probability.flattened(), axis: Tensor<Int32>(-1))
+            let cumsum = _Raw.cumsum(probability.flattened(), axis: Tensor<Int32>(-1))
             let threshold = Tensor<Float>(Float.random(in: 0...1))
 
             var index: Int = 0
 
             for j in 0..<cumsum.shape[0] {
-                if cumsum[j] >= threshold {
+                if cumsum[j].scalarized() >= threshold.scalarized() {
                     break
                 } else {
                     index = index + 1
@@ -113,7 +113,7 @@ public class KMeans {
     ///   - data: Data with shape `[sample count, feature count]`.
     internal func randomInitializer(_ data: Tensor<Float>) {
         // shuffle the input data.
-        let shuffled = Raw.randomShuffle(value: data, seed: self.seed)
+        let shuffled = _Raw.randomShuffle(value: data, seed: self.seed)
         for i in 0..<clusterCount {
             self.centroids[i] = shuffled[i]
         }

Sources/swiftML/KNeighborsClassifier.swift

@@ -113,13 +113,13 @@ public class KNeighborsClassifier {
 
         // Find the top neighbor with minimum distance.
         (minDistances, minDistanceIndex) =
-            Raw.topKV2(distances, k: Tensor<Int32>(Int32(data.shape[0])), sorted: true)
-        minDistances = Raw.reverse(minDistances, dims: Tensor<Bool>([true]))
+            _Raw.topKV2(distances, k: Tensor<Int32>(Int32(data.shape[0])), sorted: true)
+        minDistances = _Raw.reverse(minDistances, dims: Tensor<Bool>([true]))
         minDistances = minDistances
             .slice(lowerBounds: Tensor<Int32>([0]),
                 sizes: Tensor<Int32>([Int32(self.neighborCount)]))
 
-        minDistanceIndex = Raw.reverse(minDistanceIndex, dims: Tensor<Bool>([true]))
+        minDistanceIndex = _Raw.reverse(minDistanceIndex, dims: Tensor<Bool>([true]))
         minDistanceIndex = minDistanceIndex
             .slice(lowerBounds: Tensor<Int32>([0]),
                 sizes: Tensor<Int32>([Int32(self.neighborCount)]))
@@ -132,7 +132,7 @@ public class KNeighborsClassifier {
         let labelsAndWeightsTensor = computeWeights(
             distances: minDistances, labels: Tensor<Float>(minDistanceLabels))
 
-        (classes, indices) = Raw.unique(Tensor<Int32>(minDistanceLabels))
+        (classes, indices) = _Raw.unique(Tensor<Int32>(minDistanceLabels))
 
         var kClasses = Tensor<Int32>(zeros: [classes.shape[0]])
         var kWeights = Tensor<Float>(zeros: [classes.shape[0]])
@@ -148,7 +148,7 @@ public class KNeighborsClassifier {
         }
 
         // Returns class with highest weight.
-        let resultSet = Raw.topKV2(kWeights, k: Tensor<Int32>(1), sorted: true)
+        let resultSet = _Raw.topKV2(kWeights, k: Tensor<Int32>(1), sorted: true)
         let classIndex = Int(resultSet.indices[0].scalarized())
         return kClasses[classIndex]
     }

Sources/swiftML/KNeighborsRegressor.swift

@@ -108,13 +108,13 @@ public class KNeighborsRegressor {
 
         // Find the top neighbors with minimum distance.
         (minDistances, minDistanceIndex) =
-            Raw.topKV2(distances, k: Tensor<Int32>(Int32(data.shape[0])), sorted: true)
-        minDistances = Raw.reverse(minDistances, dims: Tensor<Bool>([true]))
+            _Raw.topKV2(distances, k: Tensor<Int32>(Int32(data.shape[0])), sorted: true)
+        minDistances = _Raw.reverse(minDistances, dims: Tensor<Bool>([true]))
         minDistances = minDistances
             .slice(lowerBounds: Tensor<Int32>([0]),
                 sizes: Tensor<Int32>([Int32(self.neighborCount)]))
 
-        minDistanceIndex = Raw.reverse(minDistanceIndex, dims: Tensor<Bool>([true]))
+        minDistanceIndex = _Raw.reverse(minDistanceIndex, dims: Tensor<Bool>([true]))
         minDistanceIndex = minDistanceIndex
             .slice(lowerBounds: Tensor<Int32>([0]),
                 sizes: Tensor<Int32>([Int32(self.neighborCount)]))

Sources/swiftML/LeastSquaresLinearRegression.swift

@@ -44,7 +44,7 @@ public class LeastSquaresLinearRegression: LinearRegression {
         }   
 
         // weights = (X^T.X)^-1.X^T.y
-        self.weights = matmul(matmul(Raw.matrixInverse(matmul(data.transposed(), data)),
+        self.weights = matmul(matmul(_Raw.matrixInverse(matmul(data.transposed(), data)),
             data.transposed()), labels)
     }
 }

Sources/swiftML/LogisticRegression.swift

@@ -61,19 +61,19 @@ public class LogisticRegression {
         }
         let sampleCount = Float(data.shape[0])
 
-        (self.classes, self.indices) = Raw.unique(labels.flattened())
+        (self.classes, self.indices) = _Raw.unique(labels.flattened())
 
         precondition(self.classes.shape[0] >= 2, "Labels must have atleast two classes")
 
         /// Loop through each class and apply one-vs-rest (OvR) scheme. 
         for i in 0..<self.classes.shape[0] {
-            let condition = Raw.equal(labels, classes[i])
+            let condition = _Raw.equal(labels, classes[i])
             let t = Tensor<Int32>(ones: [labels.shape[0], 1])
             let e = Tensor<Int32>(zeros: [labels.shape[0], 1])
 
             /// Create temparary labels for one-vs-rest scheme, based on class the selected class
             /// labeled as `1` while rest as `0`.
-            var tempLabels = Tensor<Float>(Raw.select(condition: condition, t: t, e: e))
+            var tempLabels = Tensor<Float>(_Raw.select(condition: condition, t: t, e: e))
             tempLabels = tempLabels.reshaped(to: [tempLabels.shape[0], 1])
 
             /// weights of selected class in one-vs-rests scheme.