tensorflow中使用class实现手写数字识别

tensorflow中使用class实现手写数字识别

代码示例：

#第一步，import
import tensorflow as tf #导入模块
import numpy as np
from tensorflow.keras.layers import Dense, Flatten
from tensorflow.keras import Model




#第二步，train, test
# mnist = tf.keras.datasets.mnist #导入mnist数据集
# (x_train, y_train), (x_test, y_test) = mnist.load_data() #分别配置训练集和测试集的输入和标签
path = './mnist.npz' #数据路径
f = np.load(path) #加载数据
x_train, y_train = f['x_train'], f['y_train'] #导入训练集的输入和标签
x_test, y_test = f['x_test'], f['y_test']  #导入测试集的输入和标签
f.close()

x_train, x_test = x_train/255.0, x_test/255.0 #把输入特征做归一化处理，让值在0-1之间，更容易让神经网络吸收

#第三步，model.Sequentiao()
# model = tf.keras.models.Sequential([  # model.Sequential()搭建神经网络
#     tf.keras.layers.Flatten(),  # 把数据集变成一位数组
#     tf.keras.layers.Dense(128, activation="relu"),  # 构建128个神经元，激活函数为relu的全连接层
#     tf.keras.layers.Dense(10, activation="softmax")  # 构建10个神经元，激活函数为softmax的全连接层
# ])
class mnistModel(Model):
    def __init__(self):
        super(mnistModel, self).__init__()
        self.flatten = Flatten() #搭建网络块，将输入特征拉直为一维数组
        self.d1 = Dense(128, activation = "relu")  #搭建网络块，全连接层，128个神经元，激活函数为relu,这一层命名为d1
        self.d2 = Dense(10, activation = "softmax") #搭建网络块，全连接层，10个神经元，激活函数为softmax,这一层命名为d2

    def call(self, x):
        x = self.flatten(x) #调用网络块
        x = self.d1(x) #调用网络块
        y = self.d2(x) #调用网络块
        return y

model = mnistModel()


#第四步，model.compile()
model.compile(  #model.compile()配置训练方法
    optimizer = "adam", #设置优化器为adam
    loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits = False), #损失函数为SparseCategoricalCrossentropy，from_logits为FALSE，代表输出是概率分布
    metrics = ['sparse_categorical_accuracy'] #标注网络评价指标为sparse_categorical_accuracy
)

#第五步，model.fit()
model.fit( #model.fit()用来执行训练过程
    x_train, #训练集的输入
    y_train, #训练集的标签
    batch_size = 32, #每一次喂入的数据是32
    epochs = 5, #迭代数是5
    validation_data = (x_test, y_test), #测试集的输入特征和标签
    validation_freq = 1 #测试的间隔次数
)

#第六步，model.summary()
model.summary() #输入神经网络的网络参数

结果为：

省略......

58720/60000 [============================>.] - ETA: 0s - loss: 0.0461 - sparse_categorical_accuracy: 0.9854
59136/60000 [============================>.] - ETA: 0s - loss: 0.0463 - sparse_categorical_accuracy: 0.9853
59552/60000 [============================>.] - ETA: 0s - loss: 0.0463 - sparse_categorical_accuracy: 0.9853
59968/60000 [============================>.] - ETA: 0s - loss: 0.0464 - sparse_categorical_accuracy: 0.9853
60000/60000 [==============================] - 9s 147us/sample - loss: 0.0464 - sparse_categorical_accuracy: 0.9853 - val_loss: 0.0808 - val_sparse_categorical_accuracy: 0.9751
Model: "mnist_model"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
flatten (Flatten)            multiple                  0         
_________________________________________________________________
dense (Dense)                multiple                  100480    
_________________________________________________________________
dense_1 (Dense)              multiple                  1290      
=================================================================
Total params: 101,770
Trainable params: 101,770
Non-trainable params: 0
_________________________________________________________________