tensorflow2.0实现MNIST分类

傷城~ 2021-09-28 14:46 495阅读 0赞

首先需要安装tensorflow2.0版本，这个版本与1.x的版本变化比较大，主要采用了keras的API，具体可以参考

一、下面是以jupyter notebook开发环境为例进行代码演示，jupyter notebook需要在命令行签名增加！

!pip install tensorflow-gpu==2.0.0-alpha0

二、导入使用到的库

from \_\_future\_\_ import absolute\_import, division, print\_function, unicode\_literals  
import tensorflow as tf

from tensorflow.keras.layers import Dense, Flatten, Conv2D  
from tensorflow.keras import Model

三、加载MNIST数据集

mnist = tf.keras.datasets.mnist

(x\_train, y\_train), (x\_test, y\_test) = mnist.load\_data()  
x\_train, x\_test = x\_train / 255.0, x\_test / 255.0

\# Add a channels dimension  
x\_train = x\_train\[..., tf.newaxis\]  
x\_test = x\_test\[..., tf.newaxis\]

四、切分数据集（训练数据、测试数据）

train\_ds = tf.data.Dataset.from\_tensor\_slices(  
    (x\_train, y\_train)).shuffle(10000).batch(32)  
test\_ds = tf.data.Dataset.from\_tensor\_slices((x\_test, y\_test)).batch(32)

五、建立模型

class MyModel(Model):  
  def \_\_init\_\_(self):  
    super(MyModel, self).\_\_init\_\_()  
    self.conv1 = Conv2D(32, 3, activation='relu')  
    self.flatten = Flatten()  
    self.d1 = Dense(128, activation='relu')  
    self.d2 = Dense(10, activation='softmax')

def call(self, x):  
    x = self.conv1(x)  
    x = self.flatten(x)  
    x = self.d1(x)  
    return self.d2(x)

model = MyModel()

六、定义损失函数和优化器

loss\_object = tf.keras.losses.SparseCategoricalCrossentropy()

optimizer = tf.keras.optimizers.Adam()

七、定义模型训练和测试的损失值和精确率

train\_loss = tf.keras.metrics.Mean(name='train\_loss')  
train\_accuracy = tf.keras.metrics.SparseCategoricalAccuracy(name='train\_accuracy')

test\_loss = tf.keras.metrics.Mean(name='test\_loss')  
test\_accuracy = tf.keras.metrics.SparseCategoricalAccuracy(name='test\_accuracy')

八、使用GradientTape进行模型训练

**@tf.function**  
def train\_step(images, labels):  
  with tf.GradientTape() as tape:  
    predictions = model(images)  
    loss = loss\_object(labels, predictions)  
  gradients = tape.gradient(loss, model.trainable\_variables)  
  optimizer.apply\_gradients(zip(gradients, model.trainable\_variables))

train\_loss(loss)  
  train\_accuracy(labels, predictions)

九、定义测试模型的方法

@tf.function  
def test\_step(images, labels):  
  predictions = model(images)  
  t\_loss = loss\_object(labels, predictions)

test\_loss(t\_loss)  
  test\_accuracy(labels, predictions)

十、开始训练，并且同时输出训练和测试的损失值和精确率

EPOCHS = 5

for epoch in range(EPOCHS):  
  for images, labels in train\_ds:  
    train\_step(images, labels)

for test\_images, test\_labels in test\_ds:  
    test\_step(test\_images, test\_labels)

template = 'Epoch \{\}, Loss: \{\}, Accuracy: \{\}, Test Loss: \{\}, Test Accuracy: \{\}'  
  print (template.format(epoch+1,  
                         train\_loss.result(),  
                         train\_accuracy.result()\*100,  
                         test\_loss.result(),  
                         test\_accuracy.result()\*100))