机器学习笔记（二十一）——Tensorflow 2（卷积与池化）
作者：互联网
本博客仅用于个人学习，不用于传播教学，主要是记自己能够看得懂的笔记（
学习知识来自：【吴恩达团队Tensorflow2.0实践系列课程第一课】TensorFlow2.0中基于TensorFlow2.0的人工智能、机器学习和深度学习简介及基础编程_哔哩哔哩_bilibili
果然还是大佬讲课好一点，有趣一点。
首先讲讲卷积。。。。。。算了卷积没什么好讲的，自己看去吧（卷积神经网络CNN完全指南终极版（一） - 知乎 (zhihu.com)，其实非常好理解，就是对图片进行一个过滤处理。
关于池化，也很好理解，比如把一个2*2的像素块合并为一个像素，这个像素的值可以是4个像素中的最大值，也可以是平均值。这样可以简化信息。
还有其他一些操作，我在代码中会标出。
import tensorflow as tf
import matplotlib.pyplot as plt
class myCallback(tf.keras.callbacks.Callback): #大佬给的判断Loss来退出训练的方法，我不懂∑( 口 ||
def on_epoch_end(self,epoch,logs={}):
if logs.get('accuracy')>0.9: #accuracy大于0.9，停止训练
print('\nReached 90% accuracy so cancelling training!')
self.model.stop_training=True
callback=myCallback() #调用类方法
mnist=tf.keras.datasets.fashion_mnist
(x_train,y_train),(x_test,y_test)=mnist.load_data()
x_train,x_test=x_train/255.0,x_test/255.0
x_train=x_train.reshape(60000,28,28,1) #为了输入卷积层必须改变一哈形状，当然只是多加了一层
x_test=x_test.reshape(10000,28,28,1)
model=tf.keras.Sequential([
tf.keras.layers.Conv2D(64,(3,3),activation='relu',input_shape=(28,28,1)), #卷积层，输入必须为(长度,宽度,深度)，用了64个过滤器，3*3的卷积窗口，并且去除负数
tf.keras.layers.MaxPooling2D(2,2), #池化层，2*2的池化
tf.keras.layers.Conv2D(64,(3,3),activation='relu'), #再来一层
tf.keras.layers.MaxPooling2D(2,2), #再来一层
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(128,activation='relu'),
tf.keras.layers.Dropout(0.2), #这是防止过拟合的Dropout层。可以将一些不重要的感知器丢弃。
tf.keras.layers.Dense(10,activation='softmax')
])
model.compile(
optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy']
)
print(model.summary()) #输出model的基本信息
his=model.fit(x_train,y_train,epochs=10,validation_data=(x_test,y_test),batch_size=64,callbacks=[callback])
#validation_data用于将模型在另一组数据上进行测试，可以是测试数据；batch_size是每次训练的一批数据的多少；callbacks就是召回函数，用于停止训练
plt.plot(his.epoch,his.history.get('loss'),label='loss')
plt.plot(his.epoch,his.history.get('val_loss'),label='val_loss') #画出Loss以及test data的Loss的变化
plt.show()
plt.plot(his.epoch,his.history.get('accuracy'),label='accuracy')
plt.plot(his.epoch,his.history.get('val_accuracy'),label='val_accuracy') #画出accuracy以及test data的accuracy的变化
plt.show()
model.evaluate(x_test,y_test)
得到结果：
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d (Conv2D) (None, 26, 26, 64) 640
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 13, 13, 64) 0
_________________________________________________________________
conv2d_1 (Conv2D) (None, 11, 11, 64) 36928
_________________________________________________________________
max_pooling2d_1 (MaxPooling2D) (None, 5, 5, 64) 0
_________________________________________________________________
flatten (Flatten) (None, 1600) 0
_________________________________________________________________
dense (Dense) (None, 128) 204928
_________________________________________________________________
dropout (Dropout) (None, 128) 0
_________________________________________________________________
dense_1 (Dense) (None, 10) 1290
=================================================================
Total params: 243,786
Trainable params: 243,786
Non-trainable params: 0
_________________________________________________________________
None
Epoch 1/10
2021-08-09 12:19:51.791600: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2021-08-09 12:19:51.988593: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
2021-08-09 12:19:52.644098: W tensorflow/stream_executor/gpu/redzone_allocator.cc:314] Internal: Invoking GPU asm compilation is supported on Cuda non-Windows platforms only
Relying on driver to perform ptx compilation.
Modify $PATH to customize ptxas location.
This message will be only logged once.
938/938 [==============================] - 12s 13ms/step - loss: 0.5003 - accuracy: 0.8183 - val_loss: 0.3815 - val_accuracy: 0.8588
Epoch 2/10
938/938 [==============================] - 13s 14ms/step - loss: 0.3294 - accuracy: 0.8791 - val_loss: 0.3131 - val_accuracy: 0.8867
Epoch 3/10
938/938 [==============================] - 13s 14ms/step - loss: 0.2796 - accuracy: 0.8970 - val_loss: 0.2901 - val_accuracy: 0.8944
Epoch 4/10
935/938 [============================>.] - ETA: 0s - loss: 0.2506 - accuracy: 0.9069
Reached 90% accuracy so cancelling training!
938/938 [==============================] - 13s 14ms/step - loss: 0.2507 - accuracy: 0.9069 - val_loss: 0.2686 - val_accuracy: 0.8988
313/313 [==============================] - 1s 4ms/step - loss: 0.2686 - accuracy: 0.8988
参考博客：
卷积神经网络CNN完全指南终极版（一） - 知乎 (zhihu.com)
Keras.layers.Conv2D参数详解 搭建图片分类 CNN （卷积神经网络） - 汪洋大海 – 蜗居 (woj.app)
如果MNIST数据集中有60000个图像，为什么该模型仅对1875个训练集图像进行训练? - IT屋-程序员软件开发技术分享社区 (it1352.com)
标签：loss,val,keras,卷积,test,池化,tf,Tensorflow,accuracy 来源： https://www.cnblogs.com/lunnyliu/p/15118093.html