1 import tensorflow as tf
2 import os
3 import numpy as np
4 from matplotlib import pyplot as plt
5
6
7 np.set_printoptions(threshold=np.inf)
8
9
10 mnist = tf.keras.datasets.mnist
11 (x_train, y_train), (x_test, y_test) = mnist.load_data()
12 x_train, x_test = x_train/255.0, x_test/255.0
13
14
15
16 model = tf.keras.models.Sequential([
17 tf.keras.layers.Flatten(),
18 tf.keras.layers.Dense(128, activation='relu'),
19 tf.keras.layers.Dense(10, activation='softmax')
20 ])
21
22 model.compile(optimizer='adam',
23 loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
24 metrics = ['sparse_categorical_accuracy'])
25
26
27
28 checkpoint_save_path = "./checkpoint/mnist.ckpt"
29 if os.path.exists(checkpoint_save_path + '.index'):
30 print('----------------load the model-----------------')
31 model.load_weights(checkpoint_save_path)
32
33
34
35 cp_callback = tf.keras.callbacks.ModelCheckpoint(
36 filepath=checkpoint_save_path,
37 save_weights_only=True,
38 save_best_only=True
39 )
40
41 history = model.fit(x_train, y_train, batch_size=32, epochs=5, validation_data=(x_test, y_test), validation_freq=1, callbacks=[cp_callback])
42
43 model.summary()
44
45
46 print(model.trainable_variables)
47 with open(',.weights.txt', 'w') as file:
48 for v in model.trainable_variables:
49 file.write(str(v.name) + '\n')
50 file.write(str(v.shape) + '\n')
51 file.write(str(v.numpy()) + '\n')
52
53
54
55 # 显示训练集和验证集的acc和loss曲线
56 acc = history.history['sparse_categorical_accuracy']
57 val_acc = history.history['val_sparse_categorical_accuracy']
58 loss = history.history['loss']
59 val_loss = history.history['val_loss']
60 plt.figure(figsize=(15, 5))
61 plt.subplot(1, 2, 1)
62 plt.plot(acc, label='Training Accuracy')
63 plt.plot(val_acc, label='Validation Accuracy')
64 plt.title('Training and Validation Accuracy')
65 #plt.legend()
66 plt.grid()
67
68 plt.subplot(1, 2, 2)
69 plt.plot(loss, label='Training Loss')
70 plt.plot(val_loss, label='Validation Loss')
71 plt.title('Training and Validation Loss')
72 plt.legend()
73 #plt.grid()
74 plt.show()
