0. 实验概述

（以图片中的深度手写数字识别实例二分类问题为例）

1.利用Tensorflow自动加载mnist数据集

代码：

import tensorflow as tffrom tensorflow.keras import datasets, layers, optimizers(xs,ys),_ = datasets.mnist.load_data()  # 自动下载mnist数据集print('datasets:',xs.shape,ys.shape)xs = tf.convert_to_tensor(xs,dtype=tf.float32)/255.  # 将mnist中的数据转为tensorflow格式db = tf.data.Dataset.from_tensor_slices((xs,ys)) #将下载的数据存入datasets数据集for step,(x,y) in enumerate(db):  #单个数据输出    print(step,x.shape,y,y.shape)

代码切割分析：

2. 手写数字识别体验

2.1 准备网络结构与优化器

利用Sequential模块。

#准备网络结构与优化器model = keras.Sequential([    #3层结构    layers.Dense(512,学习 activation='relu'),    layers.Dense(256, activation='relu'),    layers.Dense(10)])optimizer = optimizers.SGD(learning_rate=0.001)

2.2 计算损失函数与输出

with tf.GradientTape() as tape:            # [b, 28, 28] =>[b, 784]            x = tf.reshape(x, (-1, 28*28))            # Step1. compute output            # [b, 784] =>[b, 10]            out = model(x)            # Step2. compute loss            loss = tf.reduce_sum(tf.square(out - y)) / x.shape[0]

2.3 梯度计算与优化

# Step3. optimize and update w1, w2, w3, b1, b2, b3        grads = tape.gradient(loss, model.trainable_variables)        # w' = w - lr * grad        optimizer.apply_gradients(zip(grads, model.trainable_variables))

2.4 循环

2.5 完整代码

import  osimport  tensorflow as tffrom    tensorflow import kerasfrom    tensorflow.keras import layers, optimizers, datasetsos.environ['TF_CPP_MIN_LOG_LEVEL']='2'#数据集的加载(x, y), (x_val, y_val) = datasets.mnist.load_data() x = tf.convert_to_tensor(x, dtype=tf.float32) / 255.y = tf.convert_to_tensor(y, dtype=tf.int32)y = tf.one_hot(y, depth=10)print(x.shape, y.shape)train_dataset = tf.data.Dataset.from_tensor_slices((x, y))train_dataset = train_dataset.batch(200)  #一次加载200张图片#准备网络结构与优化器model = keras.Sequential([    #3层结构    layers.Dense(512, activation='relu'),    layers.Dense(256, activation='relu'),    layers.Dense(10)])optimizer = optimizers.SGD(learning_rate=0.001)#计算迭代def train_epoch(epoch):    # Step4.loop    for step, (x, y) in enumerate(train_dataset):        with tf.GradientTape() as tape:            # [b, 28, 28] =>[b, 784]            x = tf.reshape(x, (-1, 28*28))            # Step1. compute output            # [b, 784] =>[b, 10]            out = model(x)            # Step2. compute loss            loss = tf.reduce_sum(tf.square(out - y)) / x.shape[0]        # Step3. optimize and update w1, w2, w3, b1, b2, b3        grads = tape.gradient(loss, model.trainable_variables)        # w' = w - lr * grad        optimizer.apply_gradients(zip(grads, model.trainable_variables))        if step % 100 == 0:            print(epoch, step, 'loss:', loss.numpy())def train():     #计算迭代30次    for epoch in range(30):        train_epoch(epoch)if __name__ == '__main__':    train()

训练结果：

补充：os.environ['TF_CPP_MIN_LOG_LEVEL']

os.environ["TF_CPP_MIN_LOG_LEVEL"]的取值有四个：0，1，深度手写数字识别实例2，学习3，深度手写数字识别实例分别和log的学习四个等级挂钩：INFO，WARNING，深度手写数字识别实例ERROR，学习FATAL（重要性由左到右递增）

当os.environ["TF_CPP_MIN_LOG_LEVEL"]=0的深度手写数字识别实例时候，输出信息：INFO + WARNING + ERROR + FATAL
当os.environ["TF_CPP_MIN_LOG_LEVEL"]=1的学习时候，输出信息：WARNING + ERROR + FATAL
当os.environ["TF_CPP_MIN_LOG_LEVEL"]=2的深度手写数字识别实例时候，输出信息：ERROR + FATAL
当os.environ["TF_CPP_MIN_LOG_LEVEL"]=3的学习时候，输出信息：FATAL

深度手写数字识别实例