本指南将训练一个神经网络模型,对运动鞋和衬衫等服装图像进行分类。
1.加载Fashion MNIST数据集
# TensorFlow and tf.keras import tensorflow as tf from tensorflow import keras # Helper libraries import numpy as np import matplotlib.pyplot as plt print(tf.__version__)
讯享网
讯享网2.2.0
fashion_mnist = keras.datasets.fashion_mnist (train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data()
有6万张,28*28像素
讯享网print(train_images.shape) (60000, 28, 28)
2.画一张图
plt.imshow(train_images[0])

3.构造神经元模型
讯享网#构建模型 model=keras.Sequential([ keras.layers.Flatten(input_shape=(28,28)),#输入层,因为训练数据每一张是28*28的,所以input_shape=(28,28) keras.layers.Dense(128,activation=tf.nn.relu),#ReLu激活函数 keras.layers.Dense(10,activation=tf.nn.softmax)#输出层,有十个类别 ])
model.summary() (784+1)*128#加一是因为输入层和隐藏层都加了一个bias,相当于一个截距 (128+1)*10
讯享网Model: "sequential" _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= flatten (Flatten) (None, 784) 0 _________________________________________________________________ dense (Dense) (None, 128) _________________________________________________________________ dense_1 (Dense) (None, 10) 1290 ================================================================= Total params: 101,770 Trainable params: 101,770 Non-trainable params: 0 _________________________________________________________________ 1290
3.训练和评估模型
#为了让训练效果更好,进行数据归一化 train_images=train_images/255.0 test_images=test_images/255.0
讯享网#编译模型 model.compile(optimizer=tf.optimizers.Adam(),loss=tf.losses.sparse_categorical_crossentropy,metrics=['accuracy'])
#训练模型 model.fit(train_images,train_labels,epochs=5)
讯享网Epoch 1/5 1875/1875 [==============================] - 2s 1ms/step - loss: 0.5292 - accuracy: 0.8149 Epoch 2/5 1875/1875 [==============================] - 2s 1ms/step - loss: 0.3807 - accuracy: 0.8639 Epoch 3/5 1875/1875 [==============================] - 2s 1ms/step - loss: 0.3450 - accuracy: 0.8755 Epoch 4/5 1875/1875 [==============================] - 2s 1ms/step - loss: 0.3213 - accuracy: 0.8825 Epoch 5/5 1875/1875 [==============================] - 2s 1ms/step - loss: 0.3061 - accuracy: 0.8876 <tensorflow.python.keras.callbacks.History at 0x1d62490af10>
#评估模型 model.evaluate(test_images,test_labels)
讯享网313/313 [==============================] - 0s 1ms/step - loss: 0.3635 - accuracy: 0.8690 [0.54962, 0.29749]
predictions =model.predict(test_images)
讯享网model.predict(test_images)
array([[4.0e-06, 7.0e-09, 3.e-08, ..., 1.e-02, 1.e-06, 9.e-01], [3.e-06, 9.00e-14, 9.e-01, ..., 3.e-16, 8.e-09, 1.e-13], [1.e-05, 9.e-01, 6.e-09, ..., 5.e-15, 1.e-09, 1.e-15], ..., [5.e-04, 5.e-10, 2.e-04, ..., 2.e-07, 9.e-01, 4.e-09], [6.e-06, 9.e-01, 2.0e-07, ..., 1.e-11, 1.e-07, 5.e-10], [7.e-05, 1.e-08, 1.e-05, ..., 1.e-03, 1.0e-04, 5.e-06]], dtype=float32)
讯享网predictions[0]
array([4.0e-06, 7.0e-09, 3.e-08, 3.e-08, 1.e-07, 2.e-02, 1.e-06, 1.e-02, 1.e-06, 9.e-01], dtype=float32)
讯享网np.argmax(predictions[0])
plt.imshow(test_images[0])

4.自动终止训练

讯享网class myCallback(tf.keras.callbacks.Callback): def on_epoch_end(self,epoch,logs={
}): if(logs.get('loss')<0.4): print('loss太低了,我取消训练了') self.model.stop_training=True
callbacks=myCallback()
讯享网mnist=tf.keras.datasets.fashion_mnist (train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data() train_images=train_images/255.0 test_images=test_images/255.0 #构建模型 model=keras.Sequential([ keras.layers.Flatten(input_shape=(28,28)),#输入层,因为训练数据每一张是28*28的,所以input_shape=(28,28) keras.layers.Dense(128,activation=tf.nn.relu),#ReLu激活函数 keras.layers.Dense(10,activation=tf.nn.softmax)#输出层,有十个类别 ]) #编译模型 model.compile(optimizer=tf.optimizers.Adam(),loss=tf.losses.sparse_categorical_crossentropy,metrics=['accuracy']) #训练模型 model.fit(train_images,train_labels,epochs=5,callbacks=[callbacks])

版权声明:本文内容由互联网用户自发贡献,该文观点仅代表作者本人。本站仅提供信息存储空间服务,不拥有所有权,不承担相关法律责任。如发现本站有涉嫌侵权/违法违规的内容,请联系我们,一经查实,本站将立刻删除。
如需转载请保留出处:https://51itzy.com/kjqy/80192.html