Amazing-Python-Scripts

1
import tensorflow as tf
2
import tensorflowjs as tfjs
3
from tensorflow import keras
4

5
# load the data
6
(train_img, train_label), (test_img, test_label) = keras.datasets.mnist.load_data()
7
train_img = train_img.reshape([-1, 28, 28, 1])
8
test_img = test_img.reshape([-1, 28, 28, 1])
9
train_img = train_img/255.0
10
test_img = test_img/255.0
11
train_label = keras.utils.to_categorical(train_label)
12
test_label = keras.utils.to_categorical(test_label)
13

14
# define the model architecture
15
model = keras.Sequential([
16
    keras.layers.Conv2D(32, (5, 5), padding="same", input_shape=[28, 28, 1]),
17
    keras.layers.MaxPool2D((2, 2)),
18
    keras.layers.Conv2D(64, (5, 5), padding="same"),
19
    keras.layers.MaxPool2D((2, 2)),
20
    keras.layers.Flatten(),
21
    keras.layers.Dense(1024, activation='relu'),
22
    keras.layers.Dropout(0.2),
23
    keras.layers.Dense(10, activation='softmax')
24
])
25
model.compile(optimizer='adam', loss='categorical_crossentropy',
26
              metrics=['accuracy'])
27

28
# train the model
29
model.fit(train_img, train_label, validation_data=(
30
    test_img, test_label), epochs=10)
31
test_loss, test_acc = model.evaluate(test_img, test_label)
32
print('Test accuracy:', test_acc)
33

34
# save model as tfjs format
35
tfjs.converters.save_keras_model(model, 'models')
36