深度学习之BP算法①——底层手写实现(TensorFlow)

测试[[1,0,0],[0,1,0]]预测结果
# 数据集
data_x = [[0,0,1],[1,1,1],[1,0,1],[0,1,1]]
data_y = [[0],[1],[1],[0]]

# 测试集
test_x = [[1,0,0],[0,1,0]]

---

手写代码如下

import tensorflow as tf
import matplotlib.pyplot as plt

tf.set_random_seed(1)

# 数据集
data_x = [[0,0,1],[1,1,1],[1,0,1],[0,1,1]]
data_y = [[0],[1],[1],[0]]

# 测试集
test_x = [[1,0,0],[0,1,0]]

# 占位符
x = tf.placeholder(tf.float32,[None,3])
y = tf.placeholder(tf.float32,[None,1])

#定义模型
w1 = tf.Variable(tf.random_normal([3,5]),name='Weight1')
b1 = tf.Variable(tf.random_normal([5]),name='baci1')
w2 = tf.Variable(tf.random_normal([5,1]),name='Weight2')
b2 = tf.Variable(tf.random_normal([1]),name='baci2')

# 前向传播
z1 = tf.matmul(x,w1) + b1
a1 = tf.sigmoid(z1)
z2 = tf.matmul(a1,w2) + b2
a2 = tf.sigmoid(z2)

# 代价
cost = -tf.reduce_mean(y*tf.log(a2)+(1-y)*tf.log(1-a2))
cost_history = []

# 反向传播
dz2 = a2 - y
dw2 = tf.matmul(tf.transpose(a1),dz2) / tf.cast(tf.shape(x)[0],tf.float32)
db2 = tf.reduce_mean(dz2,0)

da1 = tf.matmul(dz2,tf.transpose(w2))
dz1 = da1*a1*(1-a1)
dw1 = tf.matmul(tf.transpose(x),dz1) / tf.cast(tf.shape(x)[0],tf.float32)
db1 = tf.reduce_mean(dz1,0)

alpha = 10e-2
update = [
    tf.assign(w2,w2-alpha*dw2),
    tf.assign(w1,w1-alpha*dw1),
    tf.assign(b2,b2-alpha*db2),
    tf.assign(b1,b1-alpha*db1),
]

# 开启会话
sess = tf.Session()
sess.run(tf.global_variables_initializer())

# 更新参数
for step in range(10001):
    cost_cvl,_ = sess.run([cost,update],feed_dict={x:data_x,y:data_y})

    if step % 500 == 0:
        print('step',step,'cost',cost_cvl)
    cost_history.append(cost_cvl)

# 画图
plt.rcParams['font.sans-serif'] = ['SimHei']
plt.rcParams['axes.unicode_minus'] = False
plt.plot(cost_history[1:])
plt.show()

# 预测
test_h = sess.run(a2,feed_dict={x:test_x})
print(test_h)

---

效果如下

step 0 cost 0.7326546
step 500 cost 0.09527467
step 1000 cost 0.032945182
step 1500 cost 0.018553516
step 2000 cost 0.012614363
step 2500 cost 0.0094508035
step 3000 cost 0.0075089647
step 3500 cost 0.006204647
step 4000 cost 0.005272277
step 4500 cost 0.004574685
step 5000 cost 0.004034303
step 5500 cost 0.0036040938
step 6000 cost 0.0032539144
step 6500 cost 0.0029636188
step 7000 cost 0.0027192552
step 7500 cost 0.0025109127
step 8000 cost 0.0023311898
step 8500 cost 0.002174723
step 9000 cost 0.0020373205
step 9500 cost 0.001915689
step 10000 cost 0.0018073446

[[0.9991935 ]
 [0.01141745]]