《PyTorch深度学习实践》（三）梯度下降算法-CSDN博客

1、优化算法

寻找使得函数达到最优值的参数

2、批量梯度下降算法

缺点：只能找到局部最优（是一种贪心的策略）

无法解决鞍点问题（梯度为0的点）

import numpy as np
import matplotlib.pyplot as plt

x_data = [1.0,2.0,3.0]
y_data = [2.0,4.0,6.0]

#设定一个初始值
w=1.0

#建立模型
def forward(x):
    return x*w

#计算损失函数
def cost(xs,ys):
    cost = 0
    for x,y in zip(xs,ys):
        y_pred = forward(x)
        cost += (y_pred - y)**2
    return cost/len(xs)

#计算梯度
def gradient(xs,ys):
    gradient = 0
    for x,y in zip(xs,ys):
        gradient += 2*x*(x*w-y)
    return gradient/len(xs)

#训练
print('Predict (before training)', 4, forward(4))
cost_list = []
for epoch in range(100):
    cost_val = cost(x_data, y_data)
    cost_list.append(cost_val)
    gradient_val = gradient(x_data, y_data)
    #学习率为0.01
    w-=0.01*gradient_val
    print("Epoch:",epoch,"w={:.2f}".format(w),"cost={:.2f}".format(cost_val))
print('Predict (after training)', 4, forward(4))

#可视化
plt.plot(cost_list)
plt.xlabel("epoch")
plt.ylabel("cost")
plt.show()

3、随机梯度下降

import numpy as np
import matplotlib.pyplot as plt

x_data = [1.0,2.0,3.0]
y_data = [2.0,4.0,6.0]

#设定一个初始值
w=1.0

#建立模型
def forward(x):
    return x*w

#计算损失函数
def loss(x,y):
    y_pred=forward(x)
    return (y-y_pred)**2

#计算梯度
def gradient(x,y):
    return 2*x*(x*w-y)

#训练
print('Predict (before training)', 4, forward(4))
loss_list = []
for epoch in range(100):
    for x,y in zip(x_data, y_data):
        gradient_val = gradient(x,y)
        #学习率为0.01
        w-=0.01*gradient_val
        l=loss(x,y)
        loss_list.append(l)
    print("Epoch:",epoch,"w={:.2f}".format(w),"loss={:.2f}".format(l))
print('Predict (after training)', 4, forward(4))

#可视化
plt.plot(loss_list)
plt.xlabel("epoch")
plt.xlim(-10,100)
plt.ylabel("cost")
plt.grid(True)
plt.show()