Scribd Logo
Upload

Welcome to Scribd!

  • 87 views

    Laboratorium Pembelajaran Ilmu Komputer Fakultas Ilmu Komputer Universitas Brawijaya

    Uploaded by

    Rebecca Octaviani
    The document describes a lab experiment using Madaline neural networks. It includes functions for the step bipolar activation, training Madaline, testing predictions, and calculating accuracy. Various logic functions are tested, including AND, OR, AND NOT, and XOR. XOR is able to be learned accurately due to Madaline having a hidden layer, allowing it to solve non-linearly separable problems. Testing on another dataset yields 100% accuracy compared to 75% for Hebb Net, due to Madaline using a non-linear decision boundary from its hidden layer.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Laboratorium Pembelajaran Ilmu Komputer Fakultas Ilmu Komputer Universitas Brawijaya

    Uploaded by

    Rebecca Octaviani
    87 views11 pages
    The document describes a lab experiment using Madaline neural networks. It includes functions for the step bipolar activation, training Madaline, testing predictions, and calculating accuracy. Various logic functions are tested, including AND, OR, AND NOT, and XOR. XOR is able to be learned accurately due to Madaline having a hidden layer, allowing it to solve non-linearly separable problems. Testing on another dataset yields 100% accuracy compared to 75% for Hebb Net, due to Madaline using a non-linear decision boundary from its hidden layer.

    Original Title

    JST

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    The document describes a lab experiment using Madaline neural networks. It includes functions for the step bipolar activation, training Madaline, testing predictions, and calculating accuracy. Various logic functions are tested, including AND, OR, AND NOT, and XOR. XOR is able to be learned accurately due to Madaline having a hidden layer, allowing it to solve non-linearly separable problems. Testing on another dataset yields 100% accuracy compared to 75% for Hebb Net, due to Madaline using a non-linear decision boundary from its hidden layer.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    87 views11 pages

    Laboratorium Pembelajaran Ilmu Komputer Fakultas Ilmu Komputer Universitas Brawijaya

    Uploaded by

    Rebecca Octaviani
    The document describes a lab experiment using Madaline neural networks. It includes functions for the step bipolar activation, training Madaline, testing predictions, and calculating accuracy. Various logic functions are tested, including AND, OR, AND NOT, and XOR. XOR is able to be learned accurately due to Madaline having a hidden layer, allowing it to solve non-linearly separable problems. Testing on another dataset yields 100% accuracy compared to 75% for Hebb Net, due to Madaline using a non-linear decision boundary from its hidden layer.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 11

    LABORATORIUM PEMBELAJARAN ILMU KOMPUTER

    FAKULTAS ILMU KOMPUTER


    UNIVERSITAS BRAWIJAYA
    BAB : MADALINE
    NAMA : REBECCA OCTAVIANI
    NIM : 185150201111039
    TANGGAL : 18/11/2020
    ASISTEN : - GUSTAVO THIODORUS
    - PANDU DEWABRATA

    A. Praktikum
    1. Buka Google Colaboratory melalui tautan ini.
    2. Tulis kode berikut ke dalam setiap cell pada notebook tersebut.
    a. Fungsi Step Bipolar

    def bipstep(y, th=0):


    return 1 if y >= th else -1

    b. Fungsi Training Madaline

    def fit(X, t, alpha=.01, max_err=.0001, max_epoch=10):


    w = np.random.rand(len(X[0])+1, 2)
    bias = np.ones((len(X), 1))
    X = np.hstack((bias, X))
    v = np.array([0.5, 0.5])
    by = 0.5

    stop = False
    epoch = 0
    while not stop and epoch < max_epoch:
    epoch += 1
    print('epoch:', epoch)
    max_dw = 0
    for x in range(len(X)):
    zin = np.matmul(X[x], w)
    z = np.array([(lambda zi : bipstep(zi))(zi) for zi in zin])
    yin = by + np.dot(z, v)
    y = bipstep(yin)

    # print('y =', y, 'vs t =', t[x], end=' ')


    if y != t[x]:
    if t[x] == 1:
    k = np.argmin(abs(zin))
    dw = alpha*(1-zin[k])*X[x]
    w[:,k] = w[:,k] + dw
    max_dw = max(abs(np.max(dw)), max_dw)
    # print('t = +1, max_dw:', max_dw)
    else:
    for k in range(len(zin)):
    if z[k] > 0:
    dw = alpha*(-1-zin[k])*X[x]
    w[:,k] = w[:,k] + dw
    max_dw = max(abs(np.max(dw)), max_dw)
    # print('t = -1, max_dw:', max_dw)
    else:
    # print('lewat')
    continue

    if max_err > max_dw:


    stop = True

    print('Bobot akhir:')
    print(w)
    return w

    c. Fungsi Testing Bipolar

    def predict(X, w):


    bias = np.ones((len(X), 1))
    X = np.hstack((bias, X))

    v = np.array([0.5, 0.5])
    by = 0.5
    Y = []

    for x in X:
    zin = np.matmul(x, w)
    z = zin.copy()
    z = np.array([(lambda zi : bipstep(zi))(zi) for zi in zin])
    yin = by + np.dot(z, v)
    y = bipstep(yin)
    Y.append(y)
    return Y

    c. Fungsi Hitung Akurasi

    def calc_accuracy(a, b):


    s = [1 if a[i] == b[i] else 0 for i in range(len(a))]
    return sum(s) / len(a)

    d. Logika AND

    train = (1, 1), (1, -1), (-1, 1), (-1, -1)


    target = 1, -1, -1, -1
    train = np.array(train)
    w = fit(train, target, max_epoch=1000)
    output = predict(train, w)
    accuracy = calc_accuracy(output, target)
    print('Output:', output)
    print('Target:', target)
    print('Accuracy:', accuracy)

    e. Logika OR

    train = (1, 1), (1, -1), (-1, 1), (-1, -1)


    target = 1, 1, 1, -1
    train = np.array(train)
    w = fit(train, target, max_epoch=1000)
    output = predict(train, w)
    accuracy = calc_accuracy(output, target)

    print('Output:', output)
    print('Target:', target)
    print('Accuracy:', accuracy)

    f. Logika AND NOT

    train = (1, 1), (1, -1), (-1, 1), (-1, -1)


    target = -1, 1, -1, -1
    train = np.array(train)
    w = fit(train, target, max_epoch=1000)
    output = predict(train, w)
    accuracy = calc_accuracy(output, target)

    print('Output:', output)
    print('Target:', target)
    print('Accuracy:', accuracy)

    g. Logika XOR

    train = (1, 1), (1, -1), (-1, 1), (-1, -1)


    target = -1, 1, 1, -1
    train = np.array(train)
    w = fit(train, target, max_epoch=1000)
    output = predict(train, w)
    accuracy = calc_accuracy(output, target)

    print('Output:', output)
    print('Target:', target)
    print('Accuracy:', accuracy)
    B. Screenshot

    a. Fungsi Step Bipolar

    b. Fungsi Training Madaline


    c. Fungsi Testing Bipolar

    d. Fungsi Hitung Akurasi

    e. Logika AND

    Output:
    f. Logika OR

    Output:
    g. Logika AND NOT

    Output:
    h. Logika XOR

    Output:
    C. Analisis
    1. Pada fungsi training, apa yang dilakukan oleh baris z = np.array([(lambda zi :
    bipstep(zi))(zi) for zi in zin]) tersebut?
    Jawab:
    Baris z = np.array([(lambda zi : bipstep(zi))(zi) for zi in zin])
    merupakan inisialisasi variabel z dengan memasukan nilai zin pada fungsi
    aktivasi step bipolar yang hasilnya dalam bentuk array.

    2. Berbeda dengan JST sebelumnya, mengapa Madaline berhasil mempelajari


    logika XOR dengan tepat? Jelaskan alasannya!
    Jawab:
    Madaline berhasil mempelajari XOR dengan tepat dikarenakan pada jaringan
    ini terdapat hidden layer yang mampu menyelesaikan masalah lebih rumit (non-
    linearly separable) meskipun proses training menjadi lebih rumit.

    3. Lakukan proses training dan testing menggunakan data berikut.


    Training: (-1, .5), (.5, .3), (1, 1.5), (3, 1.9)
    Target: (-1, -1, 1, 1)
    Berapakah akurasi yang didapatkan?
    Soal ini pernah ditanyakan pada Modul Bab 2 dan akurasi yang didapatkan saat
    itu adalah 75%. Menurut Anda apa yang membedakan Hebb Net dan Madaline
    sehingga menghasilkan akurasi yang berbeda?
    Jawab:
    Akurasi yang didapatkan adalah 100% pada percobaan dibawah menggunakan
    Madaline. Akurasi yang didapatkan pada jaringan Hebb Net dan Madaline
    berbeda karena pada jaringan Madaline terdapat hidden layer yang mampu
    menyelesaikan masalah rumit menggunakan non-linearly separable,
    sedangkan jaringan Hebb Net tidak terdapat hidden layer dan menggunakan
    linear.

    Kode program:
    Output:

    D. Kesimpulan

    Adaline (Adaptive Linear Neuron) adalah jaringan yang terdiri dari satu atau
    lebih input dan satu ouput. Sedangkan, Madaline (Many Adaptive Linear Neuron)
    adalah jaringan adaline yang menggunakan jaringan multilayer. Jaringan ini terdiri
    dari satu atau lebih input, satu ouput dan satu hidden layer. Maka, perbedaan kedua
    jaringan tersebut terdapat pada arsitektur yang digunakannya. Hal tersebut akan
    berpengaruh pada kedua hasil jaringan tersebut yang dimana Adaline bersifat
    linearly separable dan Madaline bersifat non-linearly separable.
    Madaline menghasilkan decision boundary (garis pemisah) yang non-linear
    karena adanya hidden layer. Jaringan tersebut hanya menghasilkan satu buah garis
    tetapi dapat memisahkan sesuai kelasnya masing-masing karena adanya garis
    pemisah yang non-linear.

    You might also like