swift-iris

Training a ML model on the Iris dataset with Swift for TensorFlow.

Requirements

• macOS 10.13.5 or later, with Xcode 10.0 or later; OR
• Ubuntu 16.04 (64-bit); OR
• other operating systems may work, but you will have to build Swift from sources.

This repository contains TensorFlow models written in Swift.

The stable branch works with the latest Swift for TensorFlow prebuilt packages.

Code

// Hyperparameters
public let inputSize = Int32(4)
public let layer1Size = Int32(10)
public let layer2Size = Int32(10)
public let outputSize = Int32(3)
let x = Tensor<Float>(shape: [batchSize, 4], scalars: featureArray)
let y_i = Tensor<Int32>(shape: [batchSize], scalars: labelArray)
var θ = IrisModel()
let η: Float = 0.001

@inlinable @inline(__always)
func lossAndGradient(_ x: Tensor<Float>, _ y_i: Tensor<Int32>, _ θ: IrisModel) -> (Float, IrisModel) {
    // Forward pass
    let z1 = x • θ.w1 + θ.b1
    let h1 = relu(z1)
    let z2 = h1 • θ.w2 + θ.b2
    let h2 = relu(z2)
    let z3 = h2 • θ.w3 + θ.b3
    let h3 = softmax(z3)
    
    // Evaluation
    let p = e_i(y_i, outputSize)
    let q = h3
    let H = -Σ(p * log(q))
    let H_total = μ(H)
    
    // Backward pass
    let dz3 = (q - p) / Float(batchSize)      // [B, l3]
    let dw3 = h2⊺ • dz3                       // [l2, l3]
    let db3 = Σ(dz3, 0)                       // [l3]
    let dz2 = dz3 • θ.w3⊺ * ajointRelu(z2)    // [B, l2]
    let dw2 = h1⊺ • dz2                       // [l1, l2]
    let db2 = Σ(dz2, 0)                       // [l2]
    let dz1 = dz2 • θ.w2⊺ * ajointRelu(z1)    // [B, l1]
    let dw1 = x⊺ • dz1                        // [x, l1]
    let db1 = Σ(dz1, 0)
    let dθ = IrisModel(w1: dw1, w2: dw2, w3: dw3, b1: db1, b2: db2, b3: db3)
    return (H_total, dθ)
}

while true {
    for i in 0..<100 {
        let (H_total, dθ) = lossAndGradient(x,y_i,θ)
        θ.update(withGradients: dθ) { θ_n, dθ_n in
            θ_n -= η * dθ_n
        }
        if i == 0 {
            // Print Loss and Likelihood
            print(String(format: "Training Loss: %.5f Likelihood: %.5f", H_total, exp(-H_total)))
        }
    }
}