梯度下降方法

先打开电脑

已于 2024-07-23 20:28:14 修改

阅读量114

点赞数 1

文章标签：计算机视觉人工智能深度学习

于 2024-07-22 21:34:00 首次发布

本文链接：https://blog.csdn.net/weixin_43710382/article/details/140619964

版权

青春练习手册专栏收录该内容

23 篇文章 0 订阅

订阅专栏

梯度下降1

# sqrt(3) = ?
# x**2 = 3 , x=?

lr = 0.01
label = 3
x = 4
epoch = 100

for e in range(100):
    print(x)
    pre = x**2
    Loss = (pre-label)**2
    delta_x = 2 * (pre - label)* 2 * x

    x = x - delta_x * lr

print(x)

梯度下降2

# (x1 - 2) ** 2 + (x2 + 3) ** 2 = 0  x1 = ? x2 = 3?
# 求解 sin（10） = ?
epoch = 10000
lr = 0.02
label = 0

x1 = 1
x2 = 1

for e in range(epoch):
    pre = (x1 - 2) ** 2 + (x2 + 3) ** 2
    loss = (pre - label) ** 2
    delta_x1 = 2 * (pre - label) *  2 * (x1 - 2)
    delta_x2 = 2 * (pre - label) * 2 * (x2 + 3)

    x1 = x1 - lr * delta_x1
    x2 = x2 - lr * delta_x2

    if e % 100 == 0:
        print(loss)
print(x1,x2)