Modul 7 Sains Data: Deep Learning dengan Keras, Regresi dan Klasifikasi Gambar

Functional API & Subclassing API dari Keras, Regresi & Klasifikasi Gambar dengan Neural Network”

Offline di Departemen Matematika
Published

May 5, 2025

Kembali ke Sains Data

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import tensorflow as tf
from tensorflow import keras

Perceptron, revisited: selain Sequential API

Di pertemuan sebelumnya, kita telah menyusun perceptron menggunakan Sequential API seperti berikut (ada dua cara yang ekuivalen):

# langsung menentukan semua layer di awal, dengan memasukkan list
model0 = keras.Sequential(
    [
        keras.layers.InputLayer(input_shape = (2,)),
        keras.layers.Dense(units = 1, activation = keras.activations.sigmoid)
    ]
)
# menambahkan layer secara berangsur-angsur
model0 = keras.Sequential()
model0.add(keras.layers.InputLayer(input_shape = (2,)))
model0.add(keras.layers.Dense(units = 1, activation = keras.activations.sigmoid))

Sequential API sebenarnya cukup terbatas: tiap layer harus berurutan satu sama lain, dan hubungan yang ada hanyalah antar pasangan dua layer yang bersebelahan.

Untuk model-model yang kita pelajari di mata kuliah Sains Data, sebenarnya Sequential API sudah cukup. Namun, kalau kalian pelajari lebih lanjut tentang neural network / deep learning, kalian akan bertemu dengan arsitektur aneh yang tidak bisa langsung disusun dengan Sequential API.

Contohnya, ada yang namanya skip connection, yaitu suatu layer terhubung dengan layer lain yang agak jauh darinya:

Sumber gambar: Aggarwal (2018) hal. 348

(Skip connection akan kalian temui kalau mempelajari residual network, yaitu arsitektur ResNet dan variasinya, yang sudah sangat di luar cakupan materi mata kuliah Sains Data.)

Untuk itu, diperlukan API selain Sequential, yaitu bisa dengan Functional API atau dengan Subclassing API. Agar kalian lebih mengenal Keras, kita akan mencoba membuat perceptron menggunakan dua API lainnya tersebut.

Kita bisa uji coba dengan dataset yang sama seperti di pertemuan sebelumnya: titik_negatif_positif.csv

df = pd.read_csv("./titik_negatif_positif.csv", dtype="float32")
inputs_df = df.drop(columns=["kelas"])
targets_df = df[["kelas"]]
inputs_arr = inputs_df.to_numpy()
targets_arr = targets_df.to_numpy()

Functional API

Ide dari Functional API adalah menyusun tiap layer dan hubungan antar layer sebagai komposisi fungsi.

Untuk Functional API, daripada keras.layers.InputLayer, gunakan keras.layers.Input

m1_input = keras.layers.Input(shape = (2,))

m1_layer1_func = keras.layers.Dense(units = 1, activation = keras.activations.sigmoid)
m1_layer1_out = m1_layer1_func(m1_input) # seperti komposisi fungsi

model1 = keras.Model(inputs=m1_input, outputs=m1_layer1_out, name="model1")
model1.summary()
Model: "model1"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ input_layer_2 (InputLayer)      │ (None, 2)              │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_2 (Dense)                 │ (None, 1)              │             3 │
└─────────────────────────────────┴────────────────────────┴───────────────┘

 Total params: 3 (12.00 B)

 Trainable params: 3 (12.00 B)

 Non-trainable params: 0 (0.00 B)

keras.utils.plot_model(
    model1,
    show_shapes = True,
    dpi=90,
    show_layer_activations = True,
    to_file = "keras_functional_model1.png"
)

Sisanya (compile lalu fit) sama dengan Sequential API

model1.compile(
    optimizer = keras.optimizers.SGD(learning_rate = 0.01),
    loss = keras.losses.BinaryCrossentropy(),
    metrics = [keras.metrics.BinaryAccuracy()]
)
history1 = model1.fit(inputs_arr, targets_arr, epochs=100, validation_split=0.2)
Epoch 1/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 1s 8ms/step - binary_accuracy: 0.5514 - loss: 0.9587 - val_binary_accuracy: 0.1800 - val_loss: 1.0349
Epoch 2/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.7520 - loss: 0.4645 - val_binary_accuracy: 0.7825 - val_loss: 0.5331
Epoch 3/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9507 - loss: 0.2794 - val_binary_accuracy: 0.9500 - val_loss: 0.3396
Epoch 4/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9894 - loss: 0.1936 - val_binary_accuracy: 0.9675 - val_loss: 0.2492
Epoch 5/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9928 - loss: 0.1603 - val_binary_accuracy: 0.9800 - val_loss: 0.1991
Epoch 6/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9951 - loss: 0.1297 - val_binary_accuracy: 0.9850 - val_loss: 0.1672
Epoch 7/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9956 - loss: 0.1105 - val_binary_accuracy: 0.9875 - val_loss: 0.1451
Epoch 8/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 5ms/step - binary_accuracy: 0.9964 - loss: 0.1003 - val_binary_accuracy: 0.9875 - val_loss: 0.1291
Epoch 9/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9976 - loss: 0.0895 - val_binary_accuracy: 0.9875 - val_loss: 0.1169
Epoch 10/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9975 - loss: 0.0810 - val_binary_accuracy: 0.9900 - val_loss: 0.1073
Epoch 11/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9983 - loss: 0.0744 - val_binary_accuracy: 0.9925 - val_loss: 0.0995
Epoch 12/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9982 - loss: 0.0744 - val_binary_accuracy: 0.9925 - val_loss: 0.0931
Epoch 13/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9985 - loss: 0.0675 - val_binary_accuracy: 0.9925 - val_loss: 0.0876
Epoch 14/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9992 - loss: 0.0587 - val_binary_accuracy: 0.9925 - val_loss: 0.0829
Epoch 15/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9977 - loss: 0.0624 - val_binary_accuracy: 0.9925 - val_loss: 0.0789
Epoch 16/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9986 - loss: 0.0596 - val_binary_accuracy: 0.9925 - val_loss: 0.0753
Epoch 17/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9958 - loss: 0.0562 - val_binary_accuracy: 0.9925 - val_loss: 0.0721
Epoch 18/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9979 - loss: 0.0493 - val_binary_accuracy: 0.9925 - val_loss: 0.0693
Epoch 19/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9979 - loss: 0.0508 - val_binary_accuracy: 0.9925 - val_loss: 0.0668
Epoch 20/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9957 - loss: 0.0512 - val_binary_accuracy: 0.9925 - val_loss: 0.0645
Epoch 21/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9988 - loss: 0.0450 - val_binary_accuracy: 0.9925 - val_loss: 0.0625
Epoch 22/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9983 - loss: 0.0457 - val_binary_accuracy: 0.9925 - val_loss: 0.0606
Epoch 23/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9980 - loss: 0.0416 - val_binary_accuracy: 0.9950 - val_loss: 0.0588
Epoch 24/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9957 - loss: 0.0431 - val_binary_accuracy: 0.9950 - val_loss: 0.0573
Epoch 25/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9992 - loss: 0.0370 - val_binary_accuracy: 0.9950 - val_loss: 0.0558
Epoch 26/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9988 - loss: 0.0380 - val_binary_accuracy: 0.9950 - val_loss: 0.0544
Epoch 27/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 5ms/step - binary_accuracy: 0.9986 - loss: 0.0379 - val_binary_accuracy: 0.9950 - val_loss: 0.0531
Epoch 28/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9983 - loss: 0.0364 - val_binary_accuracy: 0.9950 - val_loss: 0.0519
Epoch 29/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9967 - loss: 0.0364 - val_binary_accuracy: 0.9950 - val_loss: 0.0508
Epoch 30/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9976 - loss: 0.0376 - val_binary_accuracy: 0.9950 - val_loss: 0.0498
Epoch 31/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9962 - loss: 0.0364 - val_binary_accuracy: 0.9950 - val_loss: 0.0488
Epoch 32/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9987 - loss: 0.0322 - val_binary_accuracy: 0.9950 - val_loss: 0.0479
Epoch 33/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9980 - loss: 0.0309 - val_binary_accuracy: 0.9950 - val_loss: 0.0470
Epoch 34/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9982 - loss: 0.0315 - val_binary_accuracy: 0.9950 - val_loss: 0.0462
Epoch 35/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9971 - loss: 0.0294 - val_binary_accuracy: 0.9950 - val_loss: 0.0454
Epoch 36/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 6ms/step - binary_accuracy: 0.9961 - loss: 0.0327 - val_binary_accuracy: 0.9950 - val_loss: 0.0447
Epoch 37/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 1s 5ms/step - binary_accuracy: 0.9970 - loss: 0.0310 - val_binary_accuracy: 0.9950 - val_loss: 0.0440
Epoch 38/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 5ms/step - binary_accuracy: 0.9976 - loss: 0.0287 - val_binary_accuracy: 0.9950 - val_loss: 0.0433
Epoch 39/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 1s 7ms/step - binary_accuracy: 0.9972 - loss: 0.0284 - val_binary_accuracy: 0.9975 - val_loss: 0.0427
Epoch 40/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9993 - loss: 0.0259 - val_binary_accuracy: 0.9975 - val_loss: 0.0421
Epoch 41/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9982 - loss: 0.0278 - val_binary_accuracy: 0.9975 - val_loss: 0.0415
Epoch 42/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9983 - loss: 0.0288 - val_binary_accuracy: 0.9975 - val_loss: 0.0409
Epoch 43/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9995 - loss: 0.0255 - val_binary_accuracy: 0.9975 - val_loss: 0.0404
Epoch 44/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9987 - loss: 0.0273 - val_binary_accuracy: 0.9975 - val_loss: 0.0399
Epoch 45/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9978 - loss: 0.0259 - val_binary_accuracy: 0.9975 - val_loss: 0.0394
Epoch 46/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9977 - loss: 0.0257 - val_binary_accuracy: 0.9975 - val_loss: 0.0389
Epoch 47/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9979 - loss: 0.0258 - val_binary_accuracy: 0.9975 - val_loss: 0.0384
Epoch 48/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9981 - loss: 0.0272 - val_binary_accuracy: 0.9975 - val_loss: 0.0380
Epoch 49/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9966 - loss: 0.0263 - val_binary_accuracy: 0.9975 - val_loss: 0.0376
Epoch 50/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9984 - loss: 0.0237 - val_binary_accuracy: 0.9975 - val_loss: 0.0372
Epoch 51/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9964 - loss: 0.0284 - val_binary_accuracy: 0.9975 - val_loss: 0.0368
Epoch 52/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9960 - loss: 0.0256 - val_binary_accuracy: 0.9975 - val_loss: 0.0364
Epoch 53/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9977 - loss: 0.0244 - val_binary_accuracy: 0.9975 - val_loss: 0.0360
Epoch 54/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9979 - loss: 0.0276 - val_binary_accuracy: 0.9975 - val_loss: 0.0357
Epoch 55/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9971 - loss: 0.0263 - val_binary_accuracy: 0.9975 - val_loss: 0.0353
Epoch 56/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9970 - loss: 0.0254 - val_binary_accuracy: 0.9975 - val_loss: 0.0350
Epoch 57/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9971 - loss: 0.0256 - val_binary_accuracy: 0.9975 - val_loss: 0.0347
Epoch 58/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9976 - loss: 0.0238 - val_binary_accuracy: 0.9975 - val_loss: 0.0344
Epoch 59/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9983 - loss: 0.0200 - val_binary_accuracy: 0.9975 - val_loss: 0.0341
Epoch 60/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9943 - loss: 0.0304 - val_binary_accuracy: 0.9975 - val_loss: 0.0338
Epoch 61/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9992 - loss: 0.0225 - val_binary_accuracy: 0.9975 - val_loss: 0.0335
Epoch 62/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9957 - loss: 0.0253 - val_binary_accuracy: 0.9975 - val_loss: 0.0332
Epoch 63/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9975 - loss: 0.0243 - val_binary_accuracy: 0.9975 - val_loss: 0.0330
Epoch 64/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9968 - loss: 0.0233 - val_binary_accuracy: 0.9975 - val_loss: 0.0327
Epoch 65/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9981 - loss: 0.0188 - val_binary_accuracy: 0.9975 - val_loss: 0.0324
Epoch 66/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9994 - loss: 0.0195 - val_binary_accuracy: 0.9975 - val_loss: 0.0322
Epoch 67/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9992 - loss: 0.0188 - val_binary_accuracy: 0.9975 - val_loss: 0.0320
Epoch 68/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9978 - loss: 0.0215 - val_binary_accuracy: 0.9975 - val_loss: 0.0317
Epoch 69/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9938 - loss: 0.0259 - val_binary_accuracy: 0.9975 - val_loss: 0.0315
Epoch 70/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9983 - loss: 0.0214 - val_binary_accuracy: 0.9975 - val_loss: 0.0313
Epoch 71/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9973 - loss: 0.0208 - val_binary_accuracy: 0.9975 - val_loss: 0.0311
Epoch 72/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 5ms/step - binary_accuracy: 0.9953 - loss: 0.0209 - val_binary_accuracy: 0.9975 - val_loss: 0.0308
Epoch 73/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9987 - loss: 0.0194 - val_binary_accuracy: 0.9975 - val_loss: 0.0306
Epoch 74/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9977 - loss: 0.0227 - val_binary_accuracy: 0.9975 - val_loss: 0.0304
Epoch 75/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9980 - loss: 0.0205 - val_binary_accuracy: 0.9975 - val_loss: 0.0302
Epoch 76/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9987 - loss: 0.0173 - val_binary_accuracy: 0.9975 - val_loss: 0.0300
Epoch 77/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 5ms/step - binary_accuracy: 0.9990 - loss: 0.0170 - val_binary_accuracy: 0.9975 - val_loss: 0.0299
Epoch 78/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 6ms/step - binary_accuracy: 0.9959 - loss: 0.0232 - val_binary_accuracy: 0.9975 - val_loss: 0.0297
Epoch 79/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 1s 5ms/step - binary_accuracy: 0.9938 - loss: 0.0242 - val_binary_accuracy: 0.9975 - val_loss: 0.0295
Epoch 80/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 1s 8ms/step - binary_accuracy: 0.9958 - loss: 0.0220 - val_binary_accuracy: 0.9975 - val_loss: 0.0293
Epoch 81/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9975 - loss: 0.0191 - val_binary_accuracy: 0.9975 - val_loss: 0.0291
Epoch 82/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9973 - loss: 0.0208 - val_binary_accuracy: 0.9975 - val_loss: 0.0290
Epoch 83/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9987 - loss: 0.0171 - val_binary_accuracy: 0.9975 - val_loss: 0.0288
Epoch 84/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9966 - loss: 0.0221 - val_binary_accuracy: 0.9975 - val_loss: 0.0286
Epoch 85/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9973 - loss: 0.0181 - val_binary_accuracy: 0.9975 - val_loss: 0.0285
Epoch 86/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9983 - loss: 0.0183 - val_binary_accuracy: 0.9975 - val_loss: 0.0283
Epoch 87/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9957 - loss: 0.0256 - val_binary_accuracy: 0.9975 - val_loss: 0.0282
Epoch 88/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9983 - loss: 0.0193 - val_binary_accuracy: 0.9975 - val_loss: 0.0280
Epoch 89/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9981 - loss: 0.0181 - val_binary_accuracy: 0.9975 - val_loss: 0.0279
Epoch 90/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9964 - loss: 0.0187 - val_binary_accuracy: 0.9975 - val_loss: 0.0277
Epoch 91/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9982 - loss: 0.0208 - val_binary_accuracy: 0.9975 - val_loss: 0.0276
Epoch 92/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9976 - loss: 0.0164 - val_binary_accuracy: 0.9975 - val_loss: 0.0275
Epoch 93/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9962 - loss: 0.0236 - val_binary_accuracy: 0.9975 - val_loss: 0.0273
Epoch 94/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9949 - loss: 0.0219 - val_binary_accuracy: 0.9975 - val_loss: 0.0272
Epoch 95/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9977 - loss: 0.0186 - val_binary_accuracy: 0.9975 - val_loss: 0.0271
Epoch 96/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9989 - loss: 0.0159 - val_binary_accuracy: 0.9975 - val_loss: 0.0269
Epoch 97/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9986 - loss: 0.0156 - val_binary_accuracy: 0.9975 - val_loss: 0.0268
Epoch 98/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9979 - loss: 0.0196 - val_binary_accuracy: 0.9975 - val_loss: 0.0267
Epoch 99/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9986 - loss: 0.0174 - val_binary_accuracy: 0.9975 - val_loss: 0.0266
Epoch 100/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9959 - loss: 0.0201 - val_binary_accuracy: 0.9975 - val_loss: 0.0264

Kita bisa ubah dictionary .history menjadi CSV:

pd.DataFrame(history1.history).to_csv("./keras_functional_history1.csv", index=False)

Silakan download kalau mau menyocokkan/membandingkan dengan modul: keras_functional_history1.csv

Import kembali:

history1_df = pd.read_csv("./keras_functional_history1.csv")

Lalu plot loss:

plt.plot(history1_df["loss"], label = "training loss")
plt.plot(history1_df["val_loss"], label = "validation loss")
plt.xlabel("epoch")
plt.legend()
plt.show()

Subclassing API (yaitu dengan OOP)

Untuk model yang lebih kompleks, mungkin komposisi fungsi akan membuat pusing, karena banyak fungsi bertebaran di mana-mana. Agar lebih rapi dan terstruktur, kita bisa gunakan Subclassing API, yaitu dengan OOP / object oriented programming.

Silakan review Modul 2 Praktikum Struktur Data tentang Pengantar OOP kalau perlu ;)

Dalam Subclassing API, model yang kita buat berupa class yang meng-inherit (atau disebut subclassing) dari keras.Model yang sudah mengimplementasikan sebagian besar method yang kita butuhkan.

(Bahkan, kita juga bisa buat class yang hanya berupa kumpulan layer, yang nantinya akan masuk lagi ke class lain. Kalian bisa pelajari lebih lanjut: https://keras.io/guides/making_new_layers_and_models_via_subclassing/)

Dalam model yang kita susun, hanya diperlukan:

  • constructor __init__ berisi minimal satu baris, yaitu super().__init__() dan boleh berisi baris lainnya untuk menyiapkan atribut (variabel) yang langsung bisa dibuat ketika model dibuat (sebelum mulai training)

  • method call yang mendefinisikan bagaimana forward pass

  • (opsional) method build yang menyiapkan atribut yang bisa dibuat di awal training setelah ukuran input diketahui

class MyPerceptron(keras.Model):
    def __init__(self, units=1):
        super().__init__()

        # banyaknya neuron di output layer
        self.units = units

    # menyiapkan parameter (weights and biases) tergantung ukuran input
    def build(self, input_shape):
        input_dim = input_shape[-1]

        # matriks W terkadang disebut kernel
        self.kernel = self.add_weight(
            shape = (input_dim, self.units),
            initializer = keras.initializers.RandomNormal(mean=0, stddev=0.05),
            trainable = True,
        )
        self.bias = self.add_weight(
            shape = (self.units,),
            initializer = keras.initializers.RandomNormal(),
            trainable = True
        )

    # forward pass
    def call(self, inputs):
        return tf.sigmoid(
            tf.matmul(inputs, self.kernel) + self.bias
        )

Kita harus membuat instance atau objek dari class ini terlebih dahulu, lalu memanggil .build() dulu, agar kemudian bisa melakukan misalnya .fit()

model2 = MyPerceptron()
model2.build(input_shape = (2,))

Sekarang kita bisa compile, fit, simpan history, dan plot loss seperti biasa…

model2.compile(
    optimizer = keras.optimizers.SGD(learning_rate = 0.01),
    loss = keras.losses.BinaryCrossentropy(),
    metrics = [keras.metrics.BinaryAccuracy()]
)
history2 = model2.fit(inputs_arr, targets_arr, epochs=100, validation_split=0.2)
Epoch 1/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 1s 8ms/step - binary_accuracy: 0.9943 - loss: 0.5099 - val_binary_accuracy: 0.9875 - val_loss: 0.3759
Epoch 2/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9917 - loss: 0.2930 - val_binary_accuracy: 0.9875 - val_loss: 0.2786
Epoch 3/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9947 - loss: 0.2099 - val_binary_accuracy: 0.9875 - val_loss: 0.2223
Epoch 4/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9982 - loss: 0.1617 - val_binary_accuracy: 0.9900 - val_loss: 0.1860
Epoch 5/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9967 - loss: 0.1317 - val_binary_accuracy: 0.9900 - val_loss: 0.1609
Epoch 6/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9944 - loss: 0.1183 - val_binary_accuracy: 0.9900 - val_loss: 0.1426
Epoch 7/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9965 - loss: 0.1053 - val_binary_accuracy: 0.9900 - val_loss: 0.1285
Epoch 8/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9977 - loss: 0.0971 - val_binary_accuracy: 0.9900 - val_loss: 0.1175
Epoch 9/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9958 - loss: 0.0851 - val_binary_accuracy: 0.9900 - val_loss: 0.1086
Epoch 10/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9991 - loss: 0.0749 - val_binary_accuracy: 0.9900 - val_loss: 0.1011
Epoch 11/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9992 - loss: 0.0685 - val_binary_accuracy: 0.9900 - val_loss: 0.0948
Epoch 12/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9955 - loss: 0.0684 - val_binary_accuracy: 0.9900 - val_loss: 0.0895
Epoch 13/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9979 - loss: 0.0638 - val_binary_accuracy: 0.9925 - val_loss: 0.0849
Epoch 14/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9977 - loss: 0.0626 - val_binary_accuracy: 0.9925 - val_loss: 0.0808
Epoch 15/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9952 - loss: 0.0611 - val_binary_accuracy: 0.9925 - val_loss: 0.0772
Epoch 16/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9988 - loss: 0.0569 - val_binary_accuracy: 0.9925 - val_loss: 0.0741
Epoch 17/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9963 - loss: 0.0518 - val_binary_accuracy: 0.9925 - val_loss: 0.0712
Epoch 18/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9964 - loss: 0.0522 - val_binary_accuracy: 0.9925 - val_loss: 0.0687
Epoch 19/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9984 - loss: 0.0465 - val_binary_accuracy: 0.9925 - val_loss: 0.0664
Epoch 20/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9984 - loss: 0.0451 - val_binary_accuracy: 0.9925 - val_loss: 0.0643
Epoch 21/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9961 - loss: 0.0475 - val_binary_accuracy: 0.9925 - val_loss: 0.0624
Epoch 22/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9985 - loss: 0.0457 - val_binary_accuracy: 0.9925 - val_loss: 0.0606
Epoch 23/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9982 - loss: 0.0438 - val_binary_accuracy: 0.9925 - val_loss: 0.0589
Epoch 24/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9991 - loss: 0.0393 - val_binary_accuracy: 0.9925 - val_loss: 0.0574
Epoch 25/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9953 - loss: 0.0407 - val_binary_accuracy: 0.9925 - val_loss: 0.0560
Epoch 26/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9974 - loss: 0.0404 - val_binary_accuracy: 0.9925 - val_loss: 0.0547
Epoch 27/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9962 - loss: 0.0427 - val_binary_accuracy: 0.9925 - val_loss: 0.0534
Epoch 28/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9960 - loss: 0.0377 - val_binary_accuracy: 0.9925 - val_loss: 0.0523
Epoch 29/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 5ms/step - binary_accuracy: 0.9986 - loss: 0.0385 - val_binary_accuracy: 0.9925 - val_loss: 0.0512
Epoch 30/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 1s 5ms/step - binary_accuracy: 0.9980 - loss: 0.0353 - val_binary_accuracy: 0.9925 - val_loss: 0.0502
Epoch 31/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 6ms/step - binary_accuracy: 0.9974 - loss: 0.0370 - val_binary_accuracy: 0.9925 - val_loss: 0.0492
Epoch 32/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - binary_accuracy: 0.9992 - loss: 0.0324 - val_binary_accuracy: 0.9925 - val_loss: 0.0483
Epoch 33/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 1s 6ms/step - binary_accuracy: 0.9964 - loss: 0.0339 - val_binary_accuracy: 0.9925 - val_loss: 0.0475
Epoch 34/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9966 - loss: 0.0339 - val_binary_accuracy: 0.9925 - val_loss: 0.0467
Epoch 35/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9976 - loss: 0.0318 - val_binary_accuracy: 0.9925 - val_loss: 0.0459
Epoch 36/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9980 - loss: 0.0285 - val_binary_accuracy: 0.9925 - val_loss: 0.0452
Epoch 37/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9978 - loss: 0.0317 - val_binary_accuracy: 0.9925 - val_loss: 0.0445
Epoch 38/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9969 - loss: 0.0305 - val_binary_accuracy: 0.9925 - val_loss: 0.0438
Epoch 39/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9964 - loss: 0.0325 - val_binary_accuracy: 0.9925 - val_loss: 0.0432
Epoch 40/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9957 - loss: 0.0326 - val_binary_accuracy: 0.9925 - val_loss: 0.0426
Epoch 41/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9960 - loss: 0.0293 - val_binary_accuracy: 0.9925 - val_loss: 0.0420
Epoch 42/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9984 - loss: 0.0308 - val_binary_accuracy: 0.9925 - val_loss: 0.0415
Epoch 43/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9979 - loss: 0.0274 - val_binary_accuracy: 0.9925 - val_loss: 0.0409
Epoch 44/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9964 - loss: 0.0279 - val_binary_accuracy: 0.9925 - val_loss: 0.0404
Epoch 45/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9972 - loss: 0.0281 - val_binary_accuracy: 0.9925 - val_loss: 0.0399
Epoch 46/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9987 - loss: 0.0245 - val_binary_accuracy: 0.9925 - val_loss: 0.0395
Epoch 47/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9964 - loss: 0.0286 - val_binary_accuracy: 0.9925 - val_loss: 0.0390
Epoch 48/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9979 - loss: 0.0247 - val_binary_accuracy: 0.9925 - val_loss: 0.0386
Epoch 49/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9963 - loss: 0.0256 - val_binary_accuracy: 0.9925 - val_loss: 0.0381
Epoch 50/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9971 - loss: 0.0255 - val_binary_accuracy: 0.9925 - val_loss: 0.0377
Epoch 51/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9966 - loss: 0.0268 - val_binary_accuracy: 0.9925 - val_loss: 0.0373
Epoch 52/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9978 - loss: 0.0256 - val_binary_accuracy: 0.9925 - val_loss: 0.0370
Epoch 53/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9968 - loss: 0.0267 - val_binary_accuracy: 0.9925 - val_loss: 0.0366
Epoch 54/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9989 - loss: 0.0248 - val_binary_accuracy: 0.9925 - val_loss: 0.0362
Epoch 55/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9983 - loss: 0.0214 - val_binary_accuracy: 0.9925 - val_loss: 0.0359
Epoch 56/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9962 - loss: 0.0240 - val_binary_accuracy: 0.9925 - val_loss: 0.0356
Epoch 57/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9978 - loss: 0.0217 - val_binary_accuracy: 0.9925 - val_loss: 0.0353
Epoch 58/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9948 - loss: 0.0283 - val_binary_accuracy: 0.9925 - val_loss: 0.0349
Epoch 59/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9966 - loss: 0.0250 - val_binary_accuracy: 0.9925 - val_loss: 0.0346
Epoch 60/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9992 - loss: 0.0206 - val_binary_accuracy: 0.9925 - val_loss: 0.0343
Epoch 61/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9968 - loss: 0.0256 - val_binary_accuracy: 0.9925 - val_loss: 0.0340
Epoch 62/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9984 - loss: 0.0210 - val_binary_accuracy: 0.9925 - val_loss: 0.0338
Epoch 63/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9972 - loss: 0.0220 - val_binary_accuracy: 0.9925 - val_loss: 0.0335
Epoch 64/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9978 - loss: 0.0206 - val_binary_accuracy: 0.9925 - val_loss: 0.0332
Epoch 65/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9972 - loss: 0.0221 - val_binary_accuracy: 0.9925 - val_loss: 0.0330
Epoch 66/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9975 - loss: 0.0212 - val_binary_accuracy: 0.9925 - val_loss: 0.0327
Epoch 67/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9978 - loss: 0.0201 - val_binary_accuracy: 0.9925 - val_loss: 0.0325
Epoch 68/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9986 - loss: 0.0212 - val_binary_accuracy: 0.9925 - val_loss: 0.0322
Epoch 69/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9981 - loss: 0.0190 - val_binary_accuracy: 0.9925 - val_loss: 0.0320
Epoch 70/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9993 - loss: 0.0189 - val_binary_accuracy: 0.9925 - val_loss: 0.0318
Epoch 71/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9984 - loss: 0.0199 - val_binary_accuracy: 0.9925 - val_loss: 0.0316
Epoch 72/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9970 - loss: 0.0226 - val_binary_accuracy: 0.9925 - val_loss: 0.0314
Epoch 73/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9973 - loss: 0.0225 - val_binary_accuracy: 0.9950 - val_loss: 0.0312
Epoch 74/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 5ms/step - binary_accuracy: 0.9970 - loss: 0.0203 - val_binary_accuracy: 0.9950 - val_loss: 0.0309
Epoch 75/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 1s 13ms/step - binary_accuracy: 0.9968 - loss: 0.0217 - val_binary_accuracy: 0.9950 - val_loss: 0.0307
Epoch 76/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 1s 9ms/step - binary_accuracy: 0.9978 - loss: 0.0208 - val_binary_accuracy: 0.9950 - val_loss: 0.0306
Epoch 77/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 6ms/step - binary_accuracy: 0.9985 - loss: 0.0181 - val_binary_accuracy: 0.9950 - val_loss: 0.0304
Epoch 78/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9985 - loss: 0.0190 - val_binary_accuracy: 0.9950 - val_loss: 0.0302
Epoch 79/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9955 - loss: 0.0222 - val_binary_accuracy: 0.9950 - val_loss: 0.0300
Epoch 80/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9960 - loss: 0.0216 - val_binary_accuracy: 0.9950 - val_loss: 0.0298
Epoch 81/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9991 - loss: 0.0173 - val_binary_accuracy: 0.9950 - val_loss: 0.0296
Epoch 82/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9988 - loss: 0.0184 - val_binary_accuracy: 0.9950 - val_loss: 0.0295
Epoch 83/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9957 - loss: 0.0209 - val_binary_accuracy: 0.9950 - val_loss: 0.0293
Epoch 84/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9970 - loss: 0.0212 - val_binary_accuracy: 0.9950 - val_loss: 0.0291
Epoch 85/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9982 - loss: 0.0190 - val_binary_accuracy: 0.9950 - val_loss: 0.0290
Epoch 86/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9975 - loss: 0.0195 - val_binary_accuracy: 0.9950 - val_loss: 0.0288
Epoch 87/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9994 - loss: 0.0173 - val_binary_accuracy: 0.9950 - val_loss: 0.0287
Epoch 88/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9971 - loss: 0.0186 - val_binary_accuracy: 0.9950 - val_loss: 0.0285
Epoch 89/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9986 - loss: 0.0183 - val_binary_accuracy: 0.9950 - val_loss: 0.0284
Epoch 90/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9955 - loss: 0.0248 - val_binary_accuracy: 0.9950 - val_loss: 0.0282
Epoch 91/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9953 - loss: 0.0209 - val_binary_accuracy: 0.9950 - val_loss: 0.0281
Epoch 92/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9983 - loss: 0.0181 - val_binary_accuracy: 0.9950 - val_loss: 0.0279
Epoch 93/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9981 - loss: 0.0187 - val_binary_accuracy: 0.9950 - val_loss: 0.0278
Epoch 94/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9974 - loss: 0.0187 - val_binary_accuracy: 0.9950 - val_loss: 0.0277
Epoch 95/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9962 - loss: 0.0198 - val_binary_accuracy: 0.9950 - val_loss: 0.0275
Epoch 96/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9970 - loss: 0.0183 - val_binary_accuracy: 0.9950 - val_loss: 0.0274
Epoch 97/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - binary_accuracy: 0.9974 - loss: 0.0189 - val_binary_accuracy: 0.9950 - val_loss: 0.0273
Epoch 98/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9963 - loss: 0.0191 - val_binary_accuracy: 0.9950 - val_loss: 0.0271
Epoch 99/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9981 - loss: 0.0181 - val_binary_accuracy: 0.9950 - val_loss: 0.0270
Epoch 100/100
50/50 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - binary_accuracy: 0.9986 - loss: 0.0161 - val_binary_accuracy: 0.9950 - val_loss: 0.0269
pd.DataFrame(history2.history).to_csv("./keras_subclassing_history2.csv", index=False)

Silakan download kalau mau menyocokkan/membandingkan dengan modul: keras_subclassing_history2.csv

history2_df = pd.read_csv("./keras_subclassing_history2.csv")
plt.plot(history2_df["loss"], label = "training loss")
plt.plot(history2_df["val_loss"], label = "validation loss")
plt.xlabel("epoch")
plt.legend()
plt.show()

Sebenarnya, kalian bisa saja menggunakan Functional API di dalam class: siapkan fungsi-fungsinya di dalam constructor __init__ dan gunakan di dalam call

class MyPerceptron_v2(keras.Model):
    def __init__(self, units=1):
        super().__init__()

        # banyaknya neuron di output layer
        self.units = units

        # siapkan fungsi
        self.layer1_func = keras.layers.Dense(
            units = self.units,
            activation = keras.activations.sigmoid
        )

    # forward pass
    def call(self, inputs):
        x = self.layer1_func(inputs)
        return x

Contoh skip connection dengan Functional API

Kita lihat lagi gambar skip connection:

Sumber gambar: Aggarwal (2018) hal. 348

Dari gambarnya, kita bisa coba susun neural network nya:

# x
f3_input = keras.layers.Input(shape = (5,))

# weight layers
f3_layer1_func = keras.layers.Dense(units = 10, activation = keras.activations.linear)
f3_layer2_func = keras.layers.Dense(units = 5, activation = keras.activations.relu)

# F(x)
F_out = f3_layer2_func(f3_layer1_func(f3_input))

# F(x) + x
f3_layer3_out = F_out + f3_input

# membuat model akhir
model3 = keras.Model(inputs=f3_input, outputs=f3_layer3_out, name="model3")
model3.summary()
Model: "model3"

┏━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓
┃ Layer (type)        ┃ Output Shape      ┃    Param # ┃ Connected to      ┃
┡━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩
│ input_layer_3       │ (None, 5)         │          0 │ -                 │
│ (InputLayer)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ dense_3 (Dense)     │ (None, 10)        │         60 │ input_layer_3[0]… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ dense_4 (Dense)     │ (None, 5)         │         55 │ dense_3[0][0]     │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ add (Add)           │ (None, 5)         │          0 │ dense_4[0][0],    │
│                     │                   │            │ input_layer_3[0]… │
└─────────────────────┴───────────────────┴────────────┴───────────────────┘

 Total params: 115 (460.00 B)

 Trainable params: 115 (460.00 B)

 Non-trainable params: 0 (0.00 B)

keras.utils.plot_model(
    model3,
    show_shapes = True,
    dpi=90,
    show_layer_activations = True,
    to_file = "keras_functional_model3.png"
)

Apabila kode Functional API itu disusun ke dalam class, kodenya bisa menjadi seperti berikut:

class MySkipConnection(keras.Model):
    def __init__(self, units=5):
        super().__init__()

        # banyaknya neuron di output layer
        self.units = units

        # siapkan fungsi-fungsi
        self.weight1_func = keras.layers.Dense(
            units = 10,
            activation = keras.activations.linear
        )
        self.weight2_func = keras.layers.Dense(
            units = self.units,
            activation = keras.activations.relu
        )

    # forward pass
    def call(self, inputs):
        F_x = self.weight2_func(self.weight1_func(inputs))
        x = inputs
        hasil = F_x + x
        return hasil

Neural Network untuk Regresi

Ingat kembali, untuk regresi,

  • banyaknya neuron di input layer sesuai banyaknya fitur/variabel prediktor

  • banyaknya neuron di output layer sesuai banyaknya fitur/variabel target (biasanya hanya satu), dan fungsi aktivasi yang digunakan adalah fungsi aktivasi linier/identitas

  • fungsi aktivasi untuk semua hidden layer biasanya ReLU

Kita akan coba lagi dataset “California Housing Prices” (housing.csv) yang sudah kita gunakan di Modul 3 tentang regresi, yang bisa didownload dari salah satu sumber berikut:

Mari kita lihat isinya

housing_df = pd.read_csv("./housing.csv")
housing_df
longitude latitude housing_median_age total_rooms total_bedrooms population households median_income median_house_value ocean_proximity
0 -122.23 37.88 41.0 880.0 129.0 322.0 126.0 8.3252 452600.0 NEAR BAY
1 -122.22 37.86 21.0 7099.0 1106.0 2401.0 1138.0 8.3014 358500.0 NEAR BAY
2 -122.24 37.85 52.0 1467.0 190.0 496.0 177.0 7.2574 352100.0 NEAR BAY
3 -122.25 37.85 52.0 1274.0 235.0 558.0 219.0 5.6431 341300.0 NEAR BAY
4 -122.25 37.85 52.0 1627.0 280.0 565.0 259.0 3.8462 342200.0 NEAR BAY
... ... ... ... ... ... ... ... ... ... ...
20635 -121.09 39.48 25.0 1665.0 374.0 845.0 330.0 1.5603 78100.0 INLAND
20636 -121.21 39.49 18.0 697.0 150.0 356.0 114.0 2.5568 77100.0 INLAND
20637 -121.22 39.43 17.0 2254.0 485.0 1007.0 433.0 1.7000 92300.0 INLAND
20638 -121.32 39.43 18.0 1860.0 409.0 741.0 349.0 1.8672 84700.0 INLAND
20639 -121.24 39.37 16.0 2785.0 616.0 1387.0 530.0 2.3886 89400.0 INLAND

20640 rows × 10 columns

Kalau mau, kalian bisa melakukan encoding data kategorik ocean_proximity seperti di Modul 2. Tapi kali ini kita hapus/drop saja

housing_df = housing_df.drop(columns=["ocean_proximity"])
housing_df
longitude latitude housing_median_age total_rooms total_bedrooms population households median_income median_house_value
0 -122.23 37.88 41.0 880.0 129.0 322.0 126.0 8.3252 452600.0
1 -122.22 37.86 21.0 7099.0 1106.0 2401.0 1138.0 8.3014 358500.0
2 -122.24 37.85 52.0 1467.0 190.0 496.0 177.0 7.2574 352100.0
3 -122.25 37.85 52.0 1274.0 235.0 558.0 219.0 5.6431 341300.0
4 -122.25 37.85 52.0 1627.0 280.0 565.0 259.0 3.8462 342200.0
... ... ... ... ... ... ... ... ... ...
20635 -121.09 39.48 25.0 1665.0 374.0 845.0 330.0 1.5603 78100.0
20636 -121.21 39.49 18.0 697.0 150.0 356.0 114.0 2.5568 77100.0
20637 -121.22 39.43 17.0 2254.0 485.0 1007.0 433.0 1.7000 92300.0
20638 -121.32 39.43 18.0 1860.0 409.0 741.0 349.0 1.8672 84700.0
20639 -121.24 39.37 16.0 2785.0 616.0 1387.0 530.0 2.3886 89400.0

20640 rows × 9 columns

Ingat bahwa variabel target (variabel yang ingin kita prediksi) adalah median_house_value. Kita pisah dulu antara variabel prediktor (X atau inputs) dan variabel target (y atau target)

housing_X_df = housing_df.drop(columns=["median_house_value"])
housing_y_df = housing_df[["median_house_value"]]

Lalu kita ubah jadi numpy array agar bisa diolah Keras

housing_X_arr = housing_X_df.to_numpy()
housing_y_arr = housing_y_df.to_numpy()
print(housing_X_arr.shape)
print(housing_y_arr.shape)
(20640, 8)
(20640, 1)

Train test split, standarisasi:

from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(
    housing_X_arr, housing_y_arr, test_size=0.1, random_state=42
)
from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()
X_train = scaler.fit_transform(X_train)
X_test = scaler.transform(X_test)

Data target juga relatif sangat besar, sehingga sebaiknya kita scaling juga:

print(f'y min: {y_train.min()}')
print(f'y max: {y_train.max()}')
y min: 14999.0
y max: 500001.0
y_train /= 100000
y_test /= 100000
print(f'y min: {y_train.min()}')
print(f'y max: {y_train.max()}')
y min: 0.14999
y max: 5.00001

Sekarang kita bisa susun modelnya

keras.backend.clear_session()
model4 = keras.Sequential(
    [
        keras.layers.InputLayer(shape = (housing_X_arr.shape[1:])),
        keras.layers.Dense(units = 15, activation = keras.activations.relu),
        keras.layers.Dense(units = 30, activation = keras.activations.relu),
        keras.layers.Dense(units = 1, activation = keras.activations.linear)
    ]
)
model4.summary()
Model: "sequential_1"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ dense_3 (Dense)                 │ (None, 15)             │           135 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_4 (Dense)                 │ (None, 30)             │           480 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_5 (Dense)                 │ (None, 1)              │            31 │
└─────────────────────────────────┴────────────────────────┴───────────────┘

 Total params: 646 (2.52 KB)

 Trainable params: 646 (2.52 KB)

 Non-trainable params: 0 (0.00 B)

keras.utils.plot_model(
    model4,
    show_shapes = True,
    dpi=90,
    show_layer_activations = True,
    to_file = "keras_sequential_model4.png"
)

Selanjutnya, kita tentukan hyperparameter: optimizer, loss function, dan accuracy.

Ingat kembali, untuk regresi, loss function yang biasa digunakan adalah MSE (Mean Squared Error)

early_stop = keras.callbacks.EarlyStopping(
    patience=5, monitor='val_loss', restore_best_weights=True, verbose=1
)
model4.compile(
    optimizer = keras.optimizers.Adam(learning_rate = 0.001),
    loss = keras.losses.MeanSquaredError(),
    metrics = [keras.metrics.Accuracy()]
)
history4 = model4.fit(X_train, y_train, epochs=100, validation_split=1/9)
Epoch 1/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 3.5122 - val_accuracy: 0.0000e+00 - val_loss: 0.5872
Epoch 2/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.0000e+00 - loss: 0.5322 - val_accuracy: 0.0000e+00 - val_loss: 0.4665
Epoch 3/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.0000e+00 - loss: 0.4251 - val_accuracy: 0.0000e+00 - val_loss: 0.4366
Epoch 4/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.0000e+00 - loss: 0.4054 - val_accuracy: 0.0000e+00 - val_loss: 0.4219
Epoch 5/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 4ms/step - accuracy: 0.0000e+00 - loss: 0.3723 - val_accuracy: 0.0000e+00 - val_loss: 0.4157
Epoch 6/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - accuracy: 0.0000e+00 - loss: 0.3596 - val_accuracy: 0.0000e+00 - val_loss: 0.3996
Epoch 7/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.0000e+00 - loss: 0.3573 - val_accuracy: 0.0000e+00 - val_loss: 0.4091
Epoch 8/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.0000e+00 - loss: 0.3620 - val_accuracy: 0.0000e+00 - val_loss: 0.3924
Epoch 9/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.0000e+00 - loss: 0.3440 - val_accuracy: 0.0000e+00 - val_loss: 0.3816
Epoch 10/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.0000e+00 - loss: 0.3383 - val_accuracy: 0.0000e+00 - val_loss: 0.3840
Epoch 11/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.3322 - val_accuracy: 0.0000e+00 - val_loss: 0.3749
Epoch 12/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 4ms/step - accuracy: 0.0000e+00 - loss: 0.3322 - val_accuracy: 0.0000e+00 - val_loss: 0.3669
Epoch 13/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.0000e+00 - loss: 0.3320 - val_accuracy: 0.0000e+00 - val_loss: 0.3650
Epoch 14/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.0000e+00 - loss: 0.3287 - val_accuracy: 0.0000e+00 - val_loss: 0.3640
Epoch 15/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.0000e+00 - loss: 0.3248 - val_accuracy: 0.0000e+00 - val_loss: 0.3625
Epoch 16/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.3192 - val_accuracy: 0.0000e+00 - val_loss: 0.3654
Epoch 17/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.3173 - val_accuracy: 0.0000e+00 - val_loss: 0.3545
Epoch 18/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.3188 - val_accuracy: 0.0000e+00 - val_loss: 0.3528
Epoch 19/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 4ms/step - accuracy: 0.0000e+00 - loss: 0.3116 - val_accuracy: 0.0000e+00 - val_loss: 0.3555
Epoch 20/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.0000e+00 - loss: 0.3107 - val_accuracy: 0.0000e+00 - val_loss: 0.3495
Epoch 21/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.0000e+00 - loss: 0.3066 - val_accuracy: 0.0000e+00 - val_loss: 0.3530
Epoch 22/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.0000e+00 - loss: 0.3234 - val_accuracy: 0.0000e+00 - val_loss: 0.3467
Epoch 23/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.3042 - val_accuracy: 0.0000e+00 - val_loss: 0.3470
Epoch 24/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.0000e+00 - loss: 0.3006 - val_accuracy: 0.0000e+00 - val_loss: 0.3495
Epoch 25/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - accuracy: 0.0000e+00 - loss: 0.3013 - val_accuracy: 0.0000e+00 - val_loss: 0.3533
Epoch 26/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - accuracy: 0.0000e+00 - loss: 0.3031 - val_accuracy: 0.0000e+00 - val_loss: 0.3479
Epoch 27/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.3034 - val_accuracy: 0.0000e+00 - val_loss: 0.3582
Epoch 28/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.3077 - val_accuracy: 0.0000e+00 - val_loss: 0.3410
Epoch 29/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2935 - val_accuracy: 0.0000e+00 - val_loss: 0.3453
Epoch 30/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2911 - val_accuracy: 0.0000e+00 - val_loss: 0.3439
Epoch 31/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2921 - val_accuracy: 0.0000e+00 - val_loss: 0.3392
Epoch 32/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2954 - val_accuracy: 0.0000e+00 - val_loss: 0.3393
Epoch 33/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.0000e+00 - loss: 0.3028 - val_accuracy: 0.0000e+00 - val_loss: 0.3401
Epoch 34/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.0000e+00 - loss: 0.2932 - val_accuracy: 0.0000e+00 - val_loss: 0.3326
Epoch 35/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2791 - val_accuracy: 0.0000e+00 - val_loss: 0.3387
Epoch 36/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.0000e+00 - loss: 0.2888 - val_accuracy: 0.0000e+00 - val_loss: 0.3346
Epoch 37/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.0000e+00 - loss: 0.2886 - val_accuracy: 0.0000e+00 - val_loss: 0.3330
Epoch 38/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 4ms/step - accuracy: 0.0000e+00 - loss: 0.2891 - val_accuracy: 0.0000e+00 - val_loss: 0.3303
Epoch 39/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2820 - val_accuracy: 0.0000e+00 - val_loss: 0.3258
Epoch 40/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2832 - val_accuracy: 0.0000e+00 - val_loss: 0.3251
Epoch 41/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2929 - val_accuracy: 0.0000e+00 - val_loss: 0.3254
Epoch 42/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2831 - val_accuracy: 0.0000e+00 - val_loss: 0.3283
Epoch 43/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 4ms/step - accuracy: 0.0000e+00 - loss: 0.2867 - val_accuracy: 0.0000e+00 - val_loss: 0.3238
Epoch 44/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2872 - val_accuracy: 0.0000e+00 - val_loss: 0.3277
Epoch 45/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2838 - val_accuracy: 0.0000e+00 - val_loss: 0.3208
Epoch 46/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2879 - val_accuracy: 0.0000e+00 - val_loss: 0.3231
Epoch 47/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.0000e+00 - loss: 0.2849 - val_accuracy: 0.0000e+00 - val_loss: 0.3202
Epoch 48/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2874 - val_accuracy: 0.0000e+00 - val_loss: 0.3221
Epoch 49/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.0000e+00 - loss: 0.2856 - val_accuracy: 0.0000e+00 - val_loss: 0.3220
Epoch 50/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2848 - val_accuracy: 0.0000e+00 - val_loss: 0.3187
Epoch 51/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.0000e+00 - loss: 0.2791 - val_accuracy: 0.0000e+00 - val_loss: 0.3258
Epoch 52/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2770 - val_accuracy: 0.0000e+00 - val_loss: 0.3170
Epoch 53/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.0000e+00 - loss: 0.2817 - val_accuracy: 0.0000e+00 - val_loss: 0.3238
Epoch 54/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.0000e+00 - loss: 0.2719 - val_accuracy: 0.0000e+00 - val_loss: 0.3213
Epoch 55/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2789 - val_accuracy: 0.0000e+00 - val_loss: 0.3182
Epoch 56/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.0000e+00 - loss: 0.2721 - val_accuracy: 0.0000e+00 - val_loss: 0.3150
Epoch 57/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2726 - val_accuracy: 0.0000e+00 - val_loss: 0.3169
Epoch 58/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2736 - val_accuracy: 0.0000e+00 - val_loss: 0.3141
Epoch 59/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2760 - val_accuracy: 0.0000e+00 - val_loss: 0.3233
Epoch 60/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2723 - val_accuracy: 0.0000e+00 - val_loss: 0.3166
Epoch 61/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2637 - val_accuracy: 0.0000e+00 - val_loss: 0.3158
Epoch 62/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2690 - val_accuracy: 0.0000e+00 - val_loss: 0.3158
Epoch 63/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 4ms/step - accuracy: 0.0000e+00 - loss: 0.2657 - val_accuracy: 0.0000e+00 - val_loss: 0.3294
Epoch 64/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2822 - val_accuracy: 0.0000e+00 - val_loss: 0.3143
Epoch 65/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2754 - val_accuracy: 0.0000e+00 - val_loss: 0.3258
Epoch 66/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.0000e+00 - loss: 0.2616 - val_accuracy: 0.0000e+00 - val_loss: 0.3147
Epoch 67/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.0000e+00 - loss: 0.2773 - val_accuracy: 0.0000e+00 - val_loss: 0.3255
Epoch 68/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2737 - val_accuracy: 0.0000e+00 - val_loss: 0.3157
Epoch 69/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 4ms/step - accuracy: 0.0000e+00 - loss: 0.2663 - val_accuracy: 0.0000e+00 - val_loss: 0.3138
Epoch 70/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2717 - val_accuracy: 0.0000e+00 - val_loss: 0.3091
Epoch 71/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2694 - val_accuracy: 0.0000e+00 - val_loss: 0.3121
Epoch 72/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2678 - val_accuracy: 0.0000e+00 - val_loss: 0.3122
Epoch 73/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2762 - val_accuracy: 0.0000e+00 - val_loss: 0.3161
Epoch 74/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 4ms/step - accuracy: 0.0000e+00 - loss: 0.2656 - val_accuracy: 0.0000e+00 - val_loss: 0.3127
Epoch 75/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2631 - val_accuracy: 0.0000e+00 - val_loss: 0.3165
Epoch 76/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2716 - val_accuracy: 0.0000e+00 - val_loss: 0.3179
Epoch 77/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2755 - val_accuracy: 0.0000e+00 - val_loss: 0.3101
Epoch 78/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2704 - val_accuracy: 0.0000e+00 - val_loss: 0.3126
Epoch 79/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2737 - val_accuracy: 0.0000e+00 - val_loss: 0.3112
Epoch 80/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 4ms/step - accuracy: 0.0000e+00 - loss: 0.2736 - val_accuracy: 0.0000e+00 - val_loss: 0.3109
Epoch 81/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.0000e+00 - loss: 0.2713 - val_accuracy: 0.0000e+00 - val_loss: 0.3082
Epoch 82/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2843 - val_accuracy: 0.0000e+00 - val_loss: 0.3152
Epoch 83/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.0000e+00 - loss: 0.2787 - val_accuracy: 0.0000e+00 - val_loss: 0.3077
Epoch 84/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.0000e+00 - loss: 0.2691 - val_accuracy: 0.0000e+00 - val_loss: 0.3142
Epoch 85/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2593 - val_accuracy: 0.0000e+00 - val_loss: 0.3134
Epoch 86/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 4ms/step - accuracy: 0.0000e+00 - loss: 0.2697 - val_accuracy: 0.0000e+00 - val_loss: 0.3129
Epoch 87/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2602 - val_accuracy: 0.0000e+00 - val_loss: 0.3100
Epoch 88/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2681 - val_accuracy: 0.0000e+00 - val_loss: 0.3109
Epoch 89/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2630 - val_accuracy: 0.0000e+00 - val_loss: 0.3060
Epoch 90/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2712 - val_accuracy: 0.0000e+00 - val_loss: 0.3128
Epoch 91/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 4ms/step - accuracy: 0.0000e+00 - loss: 0.2644 - val_accuracy: 0.0000e+00 - val_loss: 0.3142
Epoch 92/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2726 - val_accuracy: 0.0000e+00 - val_loss: 0.3158
Epoch 93/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2741 - val_accuracy: 0.0000e+00 - val_loss: 0.3088
Epoch 94/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2726 - val_accuracy: 0.0000e+00 - val_loss: 0.3107
Epoch 95/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.0000e+00 - loss: 0.2512 - val_accuracy: 0.0000e+00 - val_loss: 0.3035
Epoch 96/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2642 - val_accuracy: 0.0000e+00 - val_loss: 0.3099
Epoch 97/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 4ms/step - accuracy: 0.0000e+00 - loss: 0.2628 - val_accuracy: 0.0000e+00 - val_loss: 0.3062
Epoch 98/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2714 - val_accuracy: 0.0000e+00 - val_loss: 0.3107
Epoch 99/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2707 - val_accuracy: 0.0000e+00 - val_loss: 0.3150
Epoch 100/100
516/516 ━━━━━━━━━━━━━━━━━━━━ 3s 3ms/step - accuracy: 0.0000e+00 - loss: 0.2713 - val_accuracy: 0.0000e+00 - val_loss: 0.3084
pd.DataFrame(history4.history).to_csv("./keras_sequential_history4.csv", index=False)

Silakan download kalau mau menyocokkan/membandingkan dengan modul: keras_sequential_history4.csv

history4_df = pd.read_csv("./keras_sequential_history4.csv")
plt.plot(history4_df["loss"], label = "training loss")
plt.plot(history4_df["val_loss"], label = "validation loss")
plt.xlabel("epoch")
plt.legend()
plt.show()

y_pred = model4.predict(X_test)

plt.hist(y_pred, color='green', alpha=.6)
plt.hist(y_test, color='blue', alpha=.6)
plt.legend(['prediction', 'truth'], loc='upper right')
plt.show()
65/65 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step

Klasifikasi Gambar dengan flatten

Gambar atau citra (image) adalah sekumpulan pixel yang disusun secara dua dimensi. Sejauh ini, neural network yang kita pelajari memiliki satu input layer yang “flat” atau datar. Sehingga, apabila kita ingin meng-input data citra ke dalam neural network, caranya adalah dengan flatten, yaitu data citra yang mula-mula dua dimensi itu disusun ulang menjadi satu dimensi.

Di Keras, ada layer istimewa untuk melakukan flatten untuk gambar berukuran a kali b pixel:

keras.layers.Flatten(input_shape = (a, b))

Ketika berurusan dengan data citra, layer ini menggantikan InputLayer yang biasa kita gunakan.

Persiapan dataset Fashion MNIST

Mari kita coba menggunakan dataset Fashion MNIST yang sudah tersedia dari Keras:

fashion_mnist = keras.datasets.fashion_mnist
(X_train_full, y_train_full), (X_test, y_test) = fashion_mnist.load_data()
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-labels-idx1-ubyte.gz
29515/29515 ━━━━━━━━━━━━━━━━━━━━ 0s 0us/step
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-images-idx3-ubyte.gz
26421880/26421880 ━━━━━━━━━━━━━━━━━━━━ 0s 0us/step
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-labels-idx1-ubyte.gz
5148/5148 ━━━━━━━━━━━━━━━━━━━━ 0s 0us/step
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-images-idx3-ubyte.gz
4422102/4422102 ━━━━━━━━━━━━━━━━━━━━ 0s 0us/step
print(f'X_train_full shape: {X_train_full.shape}')
print(f'y_train_full shape: {y_train_full.shape}')
print(f'X_test shape: {X_test.shape}')
print(f'y_test shape: {y_test.shape}')
X_train_full shape: (60000, 28, 28)
y_train_full shape: (60000,)
X_test shape: (10000, 28, 28)
y_test shape: (10000,)
X_train, X_val, y_train, y_val = train_test_split(
    X_train_full, y_train_full, test_size=1/6, random_state=42
)

print(f'X_train shape: {X_train.shape}')
print(f'y_train shape: {y_train.shape}')
print(f'X_val shape: {X_val.shape}')
print(f'y_val shape: {y_val.shape}')
print(f'X_test shape: {X_test.shape}')
print(f'y_test shape: {y_test.shape}')
X_train shape: (50000, 28, 28)
y_train shape: (50000,)
X_val shape: (10000, 28, 28)
y_val shape: (10000,)
X_test shape: (10000, 28, 28)
y_test shape: (10000,)
X_train = X_train / 255
X_val = X_val / 255
X_test = X_test / 255

Ada 10 kelas:

print(set(y_train))
{np.uint8(0), np.uint8(1), np.uint8(2), np.uint8(3), np.uint8(4), np.uint8(5), np.uint8(6), np.uint8(7), np.uint8(8), np.uint8(9)}
class_names = ["T-shirt/top", "Trouser", "Pullover", "Dress", "Coat",
               "Sandal", "Shirt", "Sneaker", "Bag", "Ankle boot"]
print(len(class_names))
10

Kita lihat salah satu gambarnya:

#@title Slider to look for some image examples {run: "auto"}
idx = 12103 #@param {type:"slider", min:0, max:49999, step:1}

plt.imshow(X_train[idx], cmap='gray')
plt.title(class_names[y_train[idx]])
plt.axis('OFF')
plt.show()

Menyusun neural network dan training

model5 = keras.Sequential(
    [
        keras.layers.Flatten(input_shape=(28,28)),
        keras.layers.Dense(units=100, activation=keras.activations.relu),
        keras.layers.Dense(units=50, activation=keras.activations.relu),
        keras.layers.Dense(units=10, activation=keras.activations.softmax)
    ]
)
model5.compile(
    optimizer = keras.optimizers.Adam(learning_rate = 0.001),
    loss = keras.losses.SparseCategoricalCrossentropy(),
    metrics = [keras.metrics.CategoricalAccuracy()]
)
model5.summary()
Model: "sequential_4"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ flatten_3 (Flatten)             │ (None, 784)            │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_12 (Dense)                │ (None, 100)            │        78,500 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_13 (Dense)                │ (None, 50)             │         5,050 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_14 (Dense)                │ (None, 10)             │           510 │
└─────────────────────────────────┴────────────────────────┴───────────────┘

 Total params: 84,060 (328.36 KB)

 Trainable params: 84,060 (328.36 KB)

 Non-trainable params: 0 (0.00 B)

history5 = model5.fit(
    X_train, y_train, validation_data=(X_val, y_val),
    epochs=50, batch_size=256
)
Epoch 1/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 3s 8ms/step - categorical_accuracy: 0.0701 - loss: 0.9699 - val_categorical_accuracy: 0.0708 - val_loss: 0.4785
Epoch 2/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 8ms/step - categorical_accuracy: 0.0557 - loss: 0.4452 - val_categorical_accuracy: 0.0696 - val_loss: 0.4183
Epoch 3/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 1s 7ms/step - categorical_accuracy: 0.0741 - loss: 0.3899 - val_categorical_accuracy: 0.0664 - val_loss: 0.3901
Epoch 4/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 3s 8ms/step - categorical_accuracy: 0.0567 - loss: 0.3662 - val_categorical_accuracy: 0.0683 - val_loss: 0.3663
Epoch 5/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 3s 13ms/step - categorical_accuracy: 0.0654 - loss: 0.3346 - val_categorical_accuracy: 0.0676 - val_loss: 0.3551
Epoch 6/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 5s 10ms/step - categorical_accuracy: 0.0714 - loss: 0.3244 - val_categorical_accuracy: 0.0687 - val_loss: 0.3429
Epoch 7/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 11ms/step - categorical_accuracy: 0.0571 - loss: 0.3054 - val_categorical_accuracy: 0.0680 - val_loss: 0.3388
Epoch 8/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 8ms/step - categorical_accuracy: 0.0583 - loss: 0.2967 - val_categorical_accuracy: 0.0691 - val_loss: 0.3395
Epoch 9/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 3s 8ms/step - categorical_accuracy: 0.0595 - loss: 0.2840 - val_categorical_accuracy: 0.0686 - val_loss: 0.3192
Epoch 10/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 3s 12ms/step - categorical_accuracy: 0.0586 - loss: 0.2682 - val_categorical_accuracy: 0.0683 - val_loss: 0.3311
Epoch 11/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 8ms/step - categorical_accuracy: 0.0674 - loss: 0.2619 - val_categorical_accuracy: 0.0688 - val_loss: 0.3219
Epoch 12/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 3s 8ms/step - categorical_accuracy: 0.0584 - loss: 0.2571 - val_categorical_accuracy: 0.0678 - val_loss: 0.3108
Epoch 13/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 8ms/step - categorical_accuracy: 0.0722 - loss: 0.2527 - val_categorical_accuracy: 0.0698 - val_loss: 0.3328
Epoch 14/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 7ms/step - categorical_accuracy: 0.0646 - loss: 0.2458 - val_categorical_accuracy: 0.0681 - val_loss: 0.3122
Epoch 15/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 1s 7ms/step - categorical_accuracy: 0.0577 - loss: 0.2397 - val_categorical_accuracy: 0.0686 - val_loss: 0.3181
Epoch 16/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 3s 10ms/step - categorical_accuracy: 0.0652 - loss: 0.2329 - val_categorical_accuracy: 0.0688 - val_loss: 0.3078
Epoch 17/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 7ms/step - categorical_accuracy: 0.0643 - loss: 0.2208 - val_categorical_accuracy: 0.0676 - val_loss: 0.3171
Epoch 18/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 3s 8ms/step - categorical_accuracy: 0.0606 - loss: 0.2227 - val_categorical_accuracy: 0.0694 - val_loss: 0.3229
Epoch 19/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 1s 7ms/step - categorical_accuracy: 0.0638 - loss: 0.2134 - val_categorical_accuracy: 0.0704 - val_loss: 0.3293
Epoch 20/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 3s 8ms/step - categorical_accuracy: 0.0658 - loss: 0.2137 - val_categorical_accuracy: 0.0678 - val_loss: 0.3176
Epoch 21/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 3s 12ms/step - categorical_accuracy: 0.0599 - loss: 0.2048 - val_categorical_accuracy: 0.0683 - val_loss: 0.3172
Epoch 22/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 8ms/step - categorical_accuracy: 0.0650 - loss: 0.1980 - val_categorical_accuracy: 0.0672 - val_loss: 0.3147
Epoch 23/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 8ms/step - categorical_accuracy: 0.0616 - loss: 0.1944 - val_categorical_accuracy: 0.0682 - val_loss: 0.3229
Epoch 24/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 3s 8ms/step - categorical_accuracy: 0.0630 - loss: 0.1895 - val_categorical_accuracy: 0.0675 - val_loss: 0.3599
Epoch 25/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 8ms/step - categorical_accuracy: 0.0631 - loss: 0.1947 - val_categorical_accuracy: 0.0691 - val_loss: 0.3424
Epoch 26/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 7ms/step - categorical_accuracy: 0.0609 - loss: 0.1853 - val_categorical_accuracy: 0.0690 - val_loss: 0.3237
Epoch 27/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 11ms/step - categorical_accuracy: 0.0690 - loss: 0.1816 - val_categorical_accuracy: 0.0691 - val_loss: 0.3355
Epoch 28/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 8ms/step - categorical_accuracy: 0.0559 - loss: 0.1784 - val_categorical_accuracy: 0.0688 - val_loss: 0.3406
Epoch 29/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 1s 7ms/step - categorical_accuracy: 0.0709 - loss: 0.1721 - val_categorical_accuracy: 0.0691 - val_loss: 0.3395
Epoch 30/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 3s 8ms/step - categorical_accuracy: 0.0650 - loss: 0.1664 - val_categorical_accuracy: 0.0692 - val_loss: 0.3663
Epoch 31/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 7ms/step - categorical_accuracy: 0.0633 - loss: 0.1687 - val_categorical_accuracy: 0.0690 - val_loss: 0.3456
Epoch 32/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 3s 9ms/step - categorical_accuracy: 0.0683 - loss: 0.1583 - val_categorical_accuracy: 0.0677 - val_loss: 0.3478
Epoch 33/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 3s 9ms/step - categorical_accuracy: 0.0703 - loss: 0.1638 - val_categorical_accuracy: 0.0684 - val_loss: 0.3615
Epoch 34/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 7ms/step - categorical_accuracy: 0.0618 - loss: 0.1530 - val_categorical_accuracy: 0.0692 - val_loss: 0.3703
Epoch 35/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 8ms/step - categorical_accuracy: 0.0676 - loss: 0.1509 - val_categorical_accuracy: 0.0673 - val_loss: 0.3660
Epoch 36/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 7ms/step - categorical_accuracy: 0.0688 - loss: 0.1524 - val_categorical_accuracy: 0.0674 - val_loss: 0.3491
Epoch 37/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 8ms/step - categorical_accuracy: 0.0684 - loss: 0.1445 - val_categorical_accuracy: 0.0688 - val_loss: 0.3403
Epoch 38/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 5s 19ms/step - categorical_accuracy: 0.0636 - loss: 0.1438 - val_categorical_accuracy: 0.0689 - val_loss: 0.3721
Epoch 39/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 3s 8ms/step - categorical_accuracy: 0.0637 - loss: 0.1415 - val_categorical_accuracy: 0.0683 - val_loss: 0.3592
Epoch 40/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 8ms/step - categorical_accuracy: 0.0617 - loss: 0.1399 - val_categorical_accuracy: 0.0686 - val_loss: 0.3710
Epoch 41/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 1s 7ms/step - categorical_accuracy: 0.0617 - loss: 0.1386 - val_categorical_accuracy: 0.0684 - val_loss: 0.3868
Epoch 42/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 3s 7ms/step - categorical_accuracy: 0.0685 - loss: 0.1320 - val_categorical_accuracy: 0.0675 - val_loss: 0.3928
Epoch 43/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 11ms/step - categorical_accuracy: 0.0664 - loss: 0.1346 - val_categorical_accuracy: 0.0682 - val_loss: 0.3956
Epoch 44/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 8ms/step - categorical_accuracy: 0.0685 - loss: 0.1229 - val_categorical_accuracy: 0.0674 - val_loss: 0.3762
Epoch 45/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 9ms/step - categorical_accuracy: 0.0620 - loss: 0.1255 - val_categorical_accuracy: 0.0696 - val_loss: 0.3801
Epoch 46/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 8ms/step - categorical_accuracy: 0.0705 - loss: 0.1233 - val_categorical_accuracy: 0.0691 - val_loss: 0.3947
Epoch 47/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 8ms/step - categorical_accuracy: 0.0719 - loss: 0.1198 - val_categorical_accuracy: 0.0688 - val_loss: 0.4037
Epoch 48/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 3s 8ms/step - categorical_accuracy: 0.0612 - loss: 0.1143 - val_categorical_accuracy: 0.0694 - val_loss: 0.4114
Epoch 49/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 3s 12ms/step - categorical_accuracy: 0.0598 - loss: 0.1191 - val_categorical_accuracy: 0.0695 - val_loss: 0.3991
Epoch 50/50
196/196 ━━━━━━━━━━━━━━━━━━━━ 2s 8ms/step - categorical_accuracy: 0.0743 - loss: 0.1145 - val_categorical_accuracy: 0.0694 - val_loss: 0.3972
pd.DataFrame(history5.history).to_csv("./keras_sequential_history5.csv", index=False)

Silakan download kalau mau menyocokkan/membandingkan dengan modul: keras_sequential_history5.csv

history5_df = pd.read_csv("./keras_sequential_history5.csv")
plt.plot(history5_df["loss"], label = "training loss")
plt.plot(history5_df["val_loss"], label = "validation loss")
plt.xlabel("epoch")
plt.legend()
plt.show()

Hasil prediksi

y_pred = model5.predict(X_test)
313/313 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step
y_pred
array([[3.89048516e-11, 4.46032236e-13, 1.15040265e-07, ...,
        7.68356957e-04, 1.53650914e-09, 9.99195278e-01],
       [1.14122786e-05, 1.14619886e-17, 9.99554217e-01, ...,
        6.40274208e-26, 1.08185377e-20, 2.92538757e-20],
       [6.54131643e-13, 9.99999940e-01, 2.02682177e-14, ...,
        3.08884612e-27, 8.57785633e-17, 7.88850808e-22],
       ...,
       [2.20215043e-07, 2.50604574e-14, 1.06251505e-08, ...,
        2.62424783e-16, 9.99996841e-01, 1.48581980e-14],
       [4.54307078e-11, 9.99999821e-01, 2.46371868e-12, ...,
        2.12386693e-15, 2.25630886e-11, 1.52572101e-13],
       [5.45784440e-09, 1.40065348e-09, 2.73625567e-08, ...,
        1.97790158e-07, 7.04383538e-08, 1.70042716e-10]], dtype=float32)
y_pred[123]
array([3.7125845e-17, 3.0286391e-20, 5.5658371e-16, 7.8263640e-24,
       6.0593184e-15, 1.0014137e-12, 1.4839321e-22, 1.1881163e-07,
       1.8030655e-16, 9.9999982e-01], dtype=float32)
np.argmax(y_pred[123])
np.int64(9)

Kita bisa melihat hasil prediksi:

#@title Slider to look for some prediction examples {run: "auto"}
idx = 4922 #@param {type:"slider", min:0, max:9999, step:1}

plt.imshow(X_test[idx], cmap='gray')
plt.title(
    f"Predicted class: {class_names[int(np.argmax(y_pred[idx]))]}\n" +
    f"True class: {class_names[y_test[idx]]}"
)
plt.axis('OFF')
plt.show()

Pengantar CNN (Convolutional Neural Network)

Sebenarnya, menerima input gambar dengan teknik flatten itu kurang efektif.

  • Dengan dense layer, bahkan dua pixel yang sangat jauh itu juga terhubungkan, padahal seharusnya tidak berhubungan.

  • Karena itu juga, tidak ada penekanan hubungan antara dua pixel yang saling berdekatan.

Alangkah baiknya, ada teknik input gambar yang bisa mempertimbangkan bagaimana hubungan suatu pixel dengan pixel-pixel di sekitarnya saja, daripada dengan semua pixel.

Convolutional Neural Network (CNN) mencoba mengatasi hal ini. Ciri khasnya adalah adanya dua jenis layer baru:

  • convolution layer

  • pooling layer, biasanya max pooling

Kedua layer baru ini bersifat sparse, yaitu beberapa neuron terhubung dengan beberapa neuron saja, tidak dengan semuanya.

Gambar berikut ini membandingkan antara sparse layer dengan dense layer:

Sumber gambar: Goodfellow, et. al. (2016) hal. 337

Konsep convolution layer

Suatu convolution layer menghitung “konvolusi” (convolution).

Sumber gambar: Kotu, hal. 325

Perhitungan konvolusi selalu melibatkan suatu “filter”, yang nilai-nilainya menjadi parameter (seperti weights and biases) yang terus di-update selama proses training.

Sumber gambar: Aggarwal (2018) hal. 321

Contoh perhitungan menggunakan filter bisa dilihat di gambar berikut.

Sumber gambar: Aggarwal (2018) hal. 336

Ketika menghitung konvolusi, filter selalu digeser. Pergeseran filter ini sebenarnya tidak harus satu langkah. Bisa saja, misalnya, dua langkah. Banyaknya langkah ini disebut stride.

Sumber gambar: Kotu, hal. 328

Konsep pooling layer

Daripada menghitung konvolusi, pooling hanya menghitung statistik sederhana saja. Biasanya menghitung maksimum, yang disebut max pooling.

Sumber gambar: Kotu, hal. 328

LeNet-5: salah satu arsitektur CNN pertama

Note: aslinya, LeNet-5 menggunakan average pooling, yaitu menghitung rata-rata, tidak seperti max pooling yang memilih maksimum.

Sumber gambar: Aggarwal (2018) hal. 41

Arsitektur LeNet-5 menggunakan Keras bisa disusun sebagai berikut:

lenet5 = keras.Sequential()

lenet5.add(keras.layers.Conv2D(
    input_shape = (32, 32, 1),
    kernel_size = (5, 5),
    filters = 6,
    activation = keras.activations.sigmoid
)) # menghasilkan C1 di gambar: ukuran 28 x 28 x 6

lenet5.add(keras.layers.AveragePooling2D(
    pool_size = (2, 2),
    strides = 2
)) # menghasilkan S2 di gambar: ukuran 14 x 14 x 6

lenet5.add(keras.layers.Conv2D(
    kernel_size = (5, 5),
    filters = 16,
    activation = keras.activations.sigmoid
)) # menghasilkan C3 di gambar: ukuran 10 x 10 x 16

lenet5.add(keras.layers.AveragePooling2D(
    pool_size = (2, 2),
    strides = 2
)) # menghasilkan S4 di gambar: ukuran 5 x 5 x 16

lenet5.add(keras.layers.Flatten())
# menjadi C5 di gambar, dengan 400 neuron

lenet5.add(keras.layers.Dense(
    units = 120, activation = keras.activations.sigmoid
))

lenet5.add(keras.layers.Dense(
    units = 84, activation = keras.activations.sigmoid
))

lenet5.add(keras.layers.Dense(
    units = 10, activation = keras.activations.softmax
))
/usr/local/lib/python3.11/dist-packages/keras/src/layers/convolutional/base_conv.py:107: UserWarning: Do not pass an `input_shape`/`input_dim` argument to a layer. When using Sequential models, prefer using an `Input(shape)` object as the first layer in the model instead.
  super().__init__(activity_regularizer=activity_regularizer, **kwargs)
keras.utils.plot_model(
    lenet5,
    show_shapes = True,
    dpi=90,
    show_layer_activations = True,
    to_file = "keras_sequential_lenet5.png"
)

Referensi

Sumber gambar

  • Aggarwal, C. Charu. 2018. Neural Networks and Deep Learning: A Textbook. Edisi Pertama. Springer.

  • Goodfellow, Ian; Bengio, Yoshua; & Courville, Aaron. 2016. Deep Learning. MIT Press.

  • Kotu, Data Science Concepts and Practice