1
2

$ pip install -q tensorflow==2.0.0-beta1
$ pip install -q h5py pyyaml

1
2
3
4
5
6
7
8
9
10
11
12

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import datasets, layers, models, callbacks
from tensorflow.keras.datasets import mnist

import os
file_path = os.path.abspath('./mnist.npz')

(train_x, train_y), (test_x, test_y) = datasets.mnist.load_data(path=file_path)
train_y, test_y = train_y[:1000], test_y[:1000]
train_x = train_x[:1000].reshape(-1, 28 * 28) / 255.0
test_x = test_x[:1000].reshape(-1, 28 * 28) / 255.0

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

def create_model():
    model = models.Sequential([
        layers.Dense(512, activation='relu', input_shape=(784,)),
        layers.Dropout(0.2),
        layers.Dense(10, activation='softmax')
    ])

    model.compile(optimizer='adam', metrics=['accuracy'],
                  loss='sparse_categorical_crossentropy')

    return model

def evaluate(target_model):
    _, acc = target_model.evaluate(test_x, test_y)
    print("Restore model, accuracy: {:5.2f}%".format(100*acc))

1
2
3
4
5
6
7
8
9
10
11

checkpoint_path = "training_2/cp-{epoch:04d}.ckpt"
checkpoint_dir = os.path.dirname(checkpoint_path)
cp_callback = callbacks.ModelCheckpoint(
    checkpoint_path, verbose=1, save_weights_only=True, period=10)

model = create_model()
model.save_weights(checkpoint_path.format(epoch=0))
model.fit(train_x, train_y, epochs=50, callbacks=[cp_callback],
          validation_data=(test_x, test_y), verbose=0)

1
2
3
4
5

Epoch 00010: saving model to training_2/cp-0010.ckpt
Epoch 00020: saving model to training_2/cp-0020.ckpt
Epoch 00030: saving model to training_2/cp-0030.ckpt
Epoch 00040: saving model to training_2/cp-0040.ckpt
Epoch 00050: saving model to training_2/cp-0050.ckpt

1
2
3
4
5

latest = tf.train.latest_checkpoint(checkpoint_dir)

model = create_model()
model.load_weights(latest)
evaluate(model)

1
2

1000/1000 [===] - 0s 90us/sample - loss: 0.4703 - accuracy: 0.8780
Restore model, accuracy: 87.80%

1
2
3
4
5

model.save_weights('./checkpoints/mannul_checkpoint')
model = create_model()
model.load_weights('./checkpoints/mannul_checkpoint')
evaluate(model)

1
2

1000/1000 [===] - 0s 90us/sample - loss: 0.4703 - accuracy: 0.8780
Restore model, accuracy: 87.80%

1

model.save('my_model.h5')

1
2

new_model = models.load_model('my_model.h5')
evaluate(new_model)

1
2

1000/1000 [===] - 0s 90us/sample - loss: 0.4703 - accuracy: 0.8780
Restore model, accuracy: 87.80%

1
2
3

import time
saved_model_path = "./saved_models/{}".format(int(time.time()))
tf.keras.experimental.export_saved_model(model, saved_model_path)

1
2

new_model = tf.keras.experimental.load_from_saved_model(saved_model_path)
model.predict(test_x).shape

1

(1000, 10)

1
2
3
4

new_model.compile(optimizer=model.optimizer,
                  loss='sparse_categorical_crossentropy',
                  metrics=['accuracy'])
evaluate(new_model)

1
2

1000/1000 [===] - 0s 90us/sample - loss: 0.4703 - accuracy: 0.8780
Restore model, accuracy: 87.80%