Tensorflow学习笔记----前向传播实战(MNIST手写数字集)

在TensorFlow中，可以通过矩阵乘法的方法，实现神经网络的前向传播过程。
a = tf.nn.relu(tf.matmul(x, w1)+b1)
y = tf.nn.relu(tf.matmul(a, w2)+b2)

在上面的代码中并没有定义w1、w2、b1、b2，TensorFlow可以通过变量(tf.Variable)，来保存和更新神经网络中的参数。比如通过下面语句可以定义w1：
weights = tf.Variable(tf.random_normal([2, 3], stddev=2))

这段代码调用了TensorFlow变量的声明函数tf.Variable。在变量声明函数中，给出了初始化这个变量的方法。

out = relu{relu{relu[X@W1+b1]@W2+b2}@W3+b3} 三层
pred = argmax(out)
loss = MSE(out,label)

在这里我们使用MNIST手写数字集来进行学习，这是一个拥有70k手写数字图片的数据集，其中60k用于训练，10k用于测试，其中彩色图片的数据表示为[28,28,3]，前向传播实现代码如下：

#前向传播
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import datasets
import os

os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'    #禁止控制台打印无关信息

# x: [60k, 28, 28],
# y: [60k]
(x, y), _ = datasets.mnist.load_data()  #加载数据集
# /255.使得x的数据范围从 [0~255] 变为 [0~1.]
x = tf.convert_to_tensor(x, dtype=tf.float32) / 255.    #创建tensor
y = tf.convert_to_tensor(y, dtype=tf.int32)

print(x.shape, y.shape, x.dtype, y.dtype)   #查看x和y的shape以及数据类型
print(tf.reduce_min(x), tf.reduce_max(x))   #查看x和y的最大、最小值
print(tf.reduce_min(y), tf.reduce_max(y))


train_db = tf.data.Dataset.from_tensor_slices((x,y)).batch(128)
train_iter = iter(train_db)     #迭代器
sample = next(train_iter)
print('batch:', sample[0].shape, sample[1].shape)


# [b, 784] => [b, 256] => [b, 128] => [b, 10]
# [dim_in, dim_out], [dim_out]
w1 = tf.Variable(tf.random.truncated_normal([784, 256], stddev=0.1))#随机
b1 = tf.Variable(tf.zeros([256]))
w2 = tf.Variable(tf.random.truncated_normal([256, 128], stddev=0.1))
b2 = tf.Variable(tf.zeros([128]))
w3 = tf.Variable(tf.random.truncated_normal([128, 10], stddev=0.1))
b3 = tf.Variable(tf.zeros([10]))

lr = 1e-3   #设置学习率为：10^-3，即0.001

for epoch in range(5): # 对数据集进行迭代
    for step, (x, y) in enumerate(train_db): # for every batch
        # x:[128, 28, 28]
        # y: [128]

        # [b, 28, 28] => [b, 28*28]
        x = tf.reshape(x, [-1, 28*28])  #维度变换

        with tf.GradientTape() as tape: # tf.Variable
            # x: [b, 28*28]
            # h1 = x@w1 + b1
            # [b, 784]@[784, 256] + [256] => [b, 256] + [256] => [b, 256] + [b, 256]
            h1 = x@w1 + tf.broadcast_to(b1, [x.shape[0], 256])
            h1 = tf.nn.relu(h1) #去线性化
            # [b, 256] => [b, 128]
            h2 = h1@w2 + b2
            h2 = tf.nn.relu(h2)
            # [b, 128] => [b, 10]
            out = h2@w3 + b3

            # compute loss计算误差
            # out: [b, 10]
            # y: [b] => [b, 10]
            y_onehot = tf.one_hot(y, depth=10)

            # mse = mean(sum(y-out)^2)
            # [b, 10]
            loss = tf.square(y_onehot - out)
            # mean: scalar
            loss = tf.reduce_mean(loss) #均值

        # compute gradients计算梯度
        grads = tape.gradient(loss, [w1, b1, w2, b2, w3, b3])
        # print(grads)
        # w1 = w1 - lr * w1_grad
        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))

其中数字集的.load_data()在第一次使用时需要下载，耗时较长，下载在电脑中后，后续运行就会直接调用了。

以上代码运行输出为：

(60000, 28, 28) (60000,) <dtype: 'float32'> <dtype: 'int32'>
tf.Tensor(0.0, shape=(), dtype=float32) tf.Tensor(1.0, shape=(), dtype=float32)
tf.Tensor(0, shape=(), dtype=int32) tf.Tensor(9, shape=(), dtype=int32)
batch: (128, 28, 28) (128,)
0 0 loss: 0.37443381547927856
0 100 loss: 0.20360812544822693
0 200 loss: 0.17638596892356873
0 300 loss: 0.16370470821857452
0 400 loss: 0.1687728464603424
1 0 loss: 0.1525443196296692
1 100 loss: 0.15411154925823212
1 200 loss: 0.14747974276542664
1 300 loss: 0.14006821811199188
1 400 loss: 0.1452929526567459
2 0 loss: 0.13192662596702576
2 100 loss: 0.13755102455615997
2 200 loss: 0.13076511025428772
2 300 loss: 0.125417098402977
2 400 loss: 0.12996098399162292
3 0 loss: 0.11843730509281158
3 100 loss: 0.12628236413002014
3 200 loss: 0.11908968538045883
3 300 loss: 0.11524822562932968
3 400 loss: 0.1192588210105896
4 0 loss: 0.1088164672255516
4 100 loss: 0.11793529987335205
4 200 loss: 0.11035821586847305
4 300 loss: 0.10763351619243622
4 400 loss: 0.11129333823919296
5 0 loss: 0.1015082597732544
5 100 loss: 0.11148501932621002
5 200 loss: 0.1035284772515297
5 300 loss: 0.10165484249591827
5 400 loss: 0.10515938699245453
6 0 loss: 0.09574250876903534
6 100 loss: 0.1062064915895462
6 200 loss: 0.09809093177318573
6 300 loss: 0.09678968787193298
6 400 loss: 0.10022337734699249
7 0 loss: 0.0910523310303688
7 100 loss: 0.10175584256649017
7 200 loss: 0.09358103573322296
7 300 loss: 0.09275438636541367
7 400 loss: 0.09611403942108154
8 0 loss: 0.0871553122997284
8 100 loss: 0.09793426096439362
8 200 loss: 0.08975998312234879
8 300 loss: 0.08933621644973755
8 400 loss: 0.09259170293807983
9 0 loss: 0.0838431864976883
9 100 loss: 0.09459478408098221
9 200 loss: 0.08652199804782867
9 300 loss: 0.08639708161354065
9 400 loss: 0.08955737948417664