tensorflow2使用矩阵运算模拟训练、验证过程

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import datasets
import os

os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

"""
=================准备数据=================
"""
#自动下载mnist数据集
(x, y), (x_test, y_test) = datasets.mnist.load_data()
#从numpy数据格式转换为tf.tensor
x = tf.convert_to_tensor(x, dtype=tf.float32) / 255.
y = tf.convert_to_tensor(y, dtype=tf.int32)
x_test = tf.convert_to_tensor(x_test, dtype=tf.float32) / 255.
y_test = tf.convert_to_tensor(y_test, dtype=tf.int32)
#将tensor转为Dataset，tf的Dataset类型可以方便的进行打散、map、batch等操作，
#这些操作简化了后续训练的数据组值过程
train_db = tf.data.Dataset.from_tensor_slices((x, y)).batch(128)
test_db = tf.data.Dataset.from_tensor_slices((x_test, y_test)).batch(128)
#train_iter = iter(train_db)
#取一批训练数据（元组类型，sample[0]为训练数据，sample[1]为标签，含128 batch）
#sample = next(train_iter)
#print('batch:', sample[0].shape, sample[1].shape)

"""
=================初始化参数=================
图像数据是28*28*1的，扁平后为784维的向量，
第一次节点数为256，第二层节点数为128，第三层节点数为10(10分类问题))
因此三个步骤的参数w1、b1、w2、b2、w3、b3的维度都必须确保可以和层的输入数据进行矩阵运算
第一层：输入数据[b, 784] @ w1 + b1 => w1为[784, 256], b1为[256] => h1为[b, 256]
第二层：输入数据[b, 256] @ w2 + b2 => w2为[256, 128], b2为[128] => h2为[b, 128]
第三层：输入数据[b, 128] @ w3 + b3 => w3为[128, 10] , b2为[10]  => out为[b, 10]
"""
#如果要让张量参与微分计算，必须是tf.Variable类型的
w1 = tf.Variable(tf.random.truncated_normal([784, 256], stddev=0.1))
b1 = tf.Variable(tf.zeros([256]))
w2 = tf.Variable(tf.random.truncated_normal([256, 128], stddev=0.1))
b2 =  tf.Variable(tf.zeros([128]))
#注意，参数初始化的过程非常重要，如果训练不收敛，可以调整一下初始化的超参，
# 如将下面的代码行中的stddev=0.1去掉，则模型不收敛。加上方差限制后，acc和loss均得到很好的训练
w3 = tf.Variable(tf.random.truncated_normal([128, 10], stddev=0.1))
b3 = tf.Variable(tf.zeros([10]))

lr = 1e-3

"""
=================开始训练=================
"""

for epoch in range(100):
    for step, (x, y) in enumerate(train_db):
        x = tf.reshape(x, [-1, 28*28])
        #记录所有在上下文中的操作，并且通过调用tfe.gradient()获得任何上下文中计算得出的张量的梯度
        with tf.GradientTape() as tape:
            h1 = x@w1 + tf.broadcast_to(b1, [x.shape[0], 256])
            h1 = tf.nn.relu(h1)
            #tf可以自动broadcast
            h2 = h1@w2 + b2
            h2 = tf.nn.relu(h2)
            out = h2@w3 + b3
            #计算loss值
            y_onehot = tf.one_hot(y, depth=10)
            # mse = mean(sum(y-out)^2)
            loss = tf.square(y_onehot - out)
            #从vactor转为scalar  向量--标量
            loss = tf.reduce_mean(loss)
        #进行微分
        grads = tape.gradient(loss, [w1, b1, w2, b2, w3, b3])
        #结合计算得到的梯度、学习率，更新参数
        w1.assign_sub(lr * grads[0])
        b1.assign_sub(lr * grads[1])
        w2.assign_sub(lr * grads[2])
        b2.assign_sub(lr * grads[3])
        w3.assign_sub(lr * grads[4])
        b3.assign_sub(lr * grads[5])

        if step % 100 == 0:
            print(epoch, step, "loss:", float(loss))

    #验证
    total_correct, total_num = 0, 0
    for step, (x, y) in enumerate(test_db):
        x = tf.reshape(x, [-1, 28*28])
        h1 = tf.nn.relu(x@w1 + b1)
        h2 = tf.nn.relu(h1@w2 + b2)
        out = h2@w3 + b3

        prob = tf.nn.softmax(out, axis=1)
        pred = tf.argmax(prob, axis=1)
        pred = tf.cast(pred, tf.int32)

        correct = tf.cast(tf.equal(pred, y), dtype=tf.int32)
        correct = tf.reduce_sum(correct)

        total_correct += int(correct)
        total_num += x.shape[0]

    acc = total_correct / total_num
    print(" test acc:", acc)

输出内容：

......
 test acc: 0.6629
21 0 loss: 0.06126173585653305
21 100 loss: 0.06827260553836823
21 200 loss: 0.06703957170248032
21 300 loss: 0.06526817381381989
21 400 loss: 0.06937036663293839
 test acc: 0.6693
22 0 loss: 0.06032110005617142
22 100 loss: 0.06739500164985657
22 200 loss: 0.0659399926662445