Add hinge loss function

Sources/TensorFlow/Loss.swift

@@ -36,6 +36,18 @@ public func meanAbsoluteError<Scalar: TensorFlowFloatingPoint>(
     return abs(expected - predicted).mean()
 }
 
+/// Returns the hinge loss between predictions and expectations.
+///
+/// - Parameters:
+///   - predicted: Predicted outputs from a neural network.
+///   - expected: Expected values, i.e. targets, that correspond to the correct output.
+@differentiable(wrt: predicted)
+public func hingeLoss<Scalar: TensorFlowFloatingPoint>(
+    predicted: Tensor<Scalar>, expected: Tensor<Scalar>
+) -> Tensor<Scalar> {
+    return max(Tensor(1) - expected * predicted, Tensor(0)).mean()
+}
+
 /// Computes the softmax cross entropy (categorical cross entropy) between logits and labels.
 ///
 /// - Parameters:

Tests/TensorFlowTests/LossTests.swift

@@ -57,6 +57,17 @@ final class LossTests: XCTestCase {
         assertElementsEqual(expected: expectedGradients, actual: gradients)
     }
 
+    func testHingeLoss() {
+        let predicted = Tensor<Float>(shape: [2, 4], scalars: [1, 2, 3, 4, 5, 6, 7, 8])
+        let expected = Tensor<Float>(
+            shape: [2, 4],
+            scalars: [0.1, 0.2, 0.3, 0.4, 0.4, 0.3, 0.2, 0.1])
+
+        let loss = hingeLoss(predicted: predicted, expected: expected)
+        let expectedLoss: Float = 0.225
+        assertElementsEqual(expected: Tensor(expectedLoss), actual: loss)
+    }
+
     func testSoftmaxCrossEntropyWithProbabilitiesLoss() {
         let logits = Tensor<Float>(shape: [2, 4], scalars: [1, 2, 3, 4, 5, 6, 7, 8])
         let labels = Tensor<Float>(