Swift for TensorFlow を使って単純なパーセプトロン

Thu 31 May 2018
swift
mani3
#swift tensorflow

Swift for TensorFlow を使って、 AND, OR, NAND, XOR を実装してみる。

ゼロから作るDeep Learning の第2章のサンプルコードを Swift for TensorFlow で書いてみただけです。

AND

func AND(a: [[Double]]) -> [Double] {
    let shape = TensorShape([Int32(a.count), Int32(a[0].count)])
    let x = Tensor(shape: shape, scalars: a.reduce([], +))
    let w = Tensor(shape: shape, repeating: 0.5)
    let b = -0.7
    let y = (x * w).sum(squeezingAxes: 1) + b
    return (y > 0).scalars.map { $0 ? 1.0 : 0.0 }
}

print(AND(a: [[0, 0], [0, 1], [1, 0], [1, 1]])) /// => [0.0, 0.0, 0.0, 1.0]

OR

func OR(a: [[Double]]) -> [Double] {
    let shape = TensorShape([Int32(a.count), Int32(a[0].count)])
    let x = Tensor(shape: shape, scalars: a.reduce([], +))
    let w = Tensor(shape: shape, repeating: 0.5)
    let b = -0.2
    let y = (x * w).sum(squeezingAxes: 1) + b
    return (y > 0).scalars.map { $0 ? 1.0 : 0.0 }
}

print(OR(a: [[0, 0], [0, 1], [1, 0], [1, 1]])) /// => [0.0, 1.0, 1.0, 1.0]

NAND

func NAND(a: [[Double]]) -> [Double] {
    let shape = TensorShape([Int32(a.count), Int32(a[0].count)])
    let x = Tensor(shape: shape, scalars: a.reduce([], +))
    let w = Tensor(shape: shape, repeating: -0.5)
    let b = 0.7
    let y = (x * w).sum(squeezingAxes: 1) + b
    return (y > 0).scalars.map { $0 ? 1.0 : 0.0 }
}

print(NAND(a: [[0, 0], [0, 1], [1, 0], [1, 1]])) /// => [1.0, 1.0, 1.0, 0.0]

XOR

func XOR(a: [[Double]]) -> [Double] {
    let s1 = NAND(a: a)
    let s2 = OR(a: a)
    return AND(a: zip(s1, s2).map { [$0, $1] })
}

print(XOR(a: [[0, 0], [0, 1], [1, 0], [1, 1]])) /// => [0.0, 1.0, 1.0, 0.0]

とりあえず出力結果があってたので大丈夫そう。何か間違ってたらごめんなさい🙇‍