导入依赖包
import tensorflow as tf
from tensorflow.keras.layers import Conv2D, BatchNormalization, Activation, MaxPool2D, Dropout, Flatten, Dense
import numpy as np
import matplotlib.pyplot as plt
print(tf.__version__)2.4.0加载数据集
(train_images, train_labels), (test_images, test_labels) = tf.keras.datasets.fashion_mnist.load_data()
print(train_images.shape, train_labels.shape)
print(test_images.shape, test_labels.shape)(60000, 28, 28) (60000,)
(10000, 28, 28) (10000,)查看训练集图片
fig, ax = plt.subplots(nrows=4, ncols=5, sharex='all', sharey='all')
ax = ax.flatten()
for i in range(20):
img = train_images[i]
ax[i].imshow(img)
ax[0].set_xticks([])
ax[0].set_yticks([])
plt.tight_layout()
plt.show()查看标签类别
labels = np.unique(train_labels)
print(labels)[0 1 2 3 4 5 6 7 8 9]对图片进行归一化
train_images, test_images = train_images / 255.0, test_images / 255.0
train_images = train_images.reshape(-1, 28, 28, 1)
test_images = test_images.reshape(-1, 28, 28, 1)
print(train_images.shape, test_images.shape)(60000, 28, 28, 1) (10000, 28, 28, 1)搭建模型
model = tf.keras.models.Sequential([
Conv2D(filters=6, kernel_size=(3, 3), padding='valid'),
BatchNormalization(),
Activation('relu'),
Conv2D(filters=6, kernel_size=(3, 3), padding='valid'),
BatchNormalization(),
Activation('relu'),
MaxPool2D(pool_size=(2, 2), strides=2, padding='same'),
Dropout(0.4),
Flatten(),
Dense(10, activation='softmax')
])指定优化器、损失函数、评价指标
model.compile(optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])训练模型
history = model.fit(train_images, train_labels,
epochs=5,
validation_data=(test_images, test_labels),
verbose=2)Epoch 1/5
1875/1875 - 44s - loss: 0.5619 - accuracy: 0.7976 - val_loss: 0.3965 - val_accuracy: 0.8561
Epoch 2/5
1875/1875 - 41s - loss: 0.4037 - accuracy: 0.8546 - val_loss: 0.3772 - val_accuracy: 0.8620
Epoch 3/5
1875/1875 - 41s - loss: 0.3713 - accuracy: 0.8656 - val_loss: 0.3355 - val_accuracy: 0.8835
Epoch 4/5
1875/1875 - 41s - loss: 0.3556 - accuracy: 0.8733 - val_loss: 0.3280 - val_accuracy: 0.8824
Epoch 5/5
1875/1875 - 48s - loss: 0.3433 - accuracy: 0.8756 - val_loss: 0.3199 - val_accuracy: 0.8871查看模型结构
model.summary()Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d (Conv2D) (None, 26, 26, 6) 60
_________________________________________________________________
batch_normalization (BatchNo (None, 26, 26, 6) 24
_________________________________________________________________
activation (Activation) (None, 26, 26, 6) 0
_________________________________________________________________
conv2d_1 (Conv2D) (None, 24, 24, 6) 330
_________________________________________________________________
batch_normalization_1 (Batch (None, 24, 24, 6) 24
_________________________________________________________________
activation_1 (Activation) (None, 24, 24, 6) 0
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 12, 12, 6) 0
_________________________________________________________________
dropout (Dropout) (None, 12, 12, 6) 0
_________________________________________________________________
flatten (Flatten) (None, 864) 0
_________________________________________________________________
dense (Dense) (None, 10) 8650
=================================================================
Total params: 9,088
Trainable params: 9,064
Non-trainable params: 24
_________________________________________________________________显示模型准确率的变化过程
plt.plot(history.epoch, history.history.get('accuracy'), label='accuracy')
plt.plot(history.epoch, history.history.get('val_accuracy'), label='val_accuracy')
plt.legend()
plt.show()显示模型损失的变化过程
plt.plot(history.epoch, history.history.get('loss'), label='loss')
plt.plot(history.epoch, history.history.get('val_loss'), label='val_loss')
plt.legend()
plt.show()
转载请注明来源，欢迎对文章中的引用来源进行考证，欢迎指出任何有错误或不够清晰的表达。可以在下面评论区评论，也可以邮件至 2621041184@qq.com