## 使用梯度下降法做线性回归

## 使用梯度下降法做线性回归

import pandas as pd
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns
from sklearn.linear_model import LinearRegression
from sklearn.metrics import mean_squared_error
X = [3,4,6,12,9,15,10,1,8,13] ## 薪水
y = [16,29,43,65,51,89,57,9,53,68]  ## 工资
# 画出薪水和工资的散点图
plt.scatter(X,y,label='salary-wage')
plt.xlabel('X')
plt.ylabel('y')
plt.show()

 

# 假设线性方程为 is y = beta * x + b, 请定义一个线性方程, 对于输入X, beta, b, 会产生输出y, y为一个列表
def line_formula(x, beta, b):
    y=[]
    for xi in x:
        yi=beta*xi+b
        y.append(yi)
    return y


# 假设X=薪水, beta = 1, b = 0, 带入上面的函数, 得到y, 画一个线图, 和薪水和工资的散点图在一张图上
y_m1 = line_formula(X, 1, 0)
plt.scatter(X,y,label='salary-wage')
plt.plot(X,y_m1,label='line')
plt.xlabel('X')
plt.ylabel('y')
plt.show()

# 计算损失函数mse 1/n*sum(y-y_out)^2
def mse(true, pred):
    n = len(true)
    mse = sum((true[i]-pred[i])**2 for i in range(n)) / n
    return mse

 

 

# 计算mse(y, y_m1)
mse_m1 = mse(y, y_m1)
mse_m1

# 计算不同的beta值的mse值, beta = range(0,9), 将不同的beta值带入到line_formula中, 画图(线图和散点图), 并打印每个bata对应的mse值, 将这些mse值保存在变量mse_values中
import pandas as pd
beta = list(range(0,9))
mse_values = []
for i in range(len(beta)):
    y_m1 = line_formula(X, beta[i], 0)
    mse_m1 = mse(y, y_m1)
    plt.scatter(X,y,label='salary-wage')
    plt.plot(X,y_m1,label='line')
    mse_values.append(mse_m1)
print(mse_values)
MSE=pd.DataFrame({'beta':beta,'MSE_':mse_values})
MSE=MSE.set_index('beta')
print(MSE)

 

# 画出损失函数的图
plt.plot(MSE.index,MSE['MSE_'],label='bata-mse')

 

# 找出mse_values最小的值
minimum = MSE['MSE_'].min()
minimum_index = MSE['MSE_'].idxmin()
print(f"mse最小值: {minimum}, 对应的beta: {minimum_index}")

 

# 定义一个简单的梯度下降, 将函数补充完整
import numpy as np
def Simple_LinearRegression_Gradient_Descent(x,y,epochs,learning_rate,n):
    m_curr = c_curr = 0
    cost_val = []
    intercept = []
    coefficient = []
    for i in range(epochs):
        y_pred = m_curr * x + c_curr
        ## 根据上面的公式分别计算损失、slope和intecept
        cost = (1/n) * sum([val**2 for val in (y-y_pred)])
        md = -(2/n) * sum(x * (y - y_pred))
        cd = -(2/n) * sum(y - y_pred)
        cost_val.append(cost)  
        m_curr = m_curr - learning_rate * md  # step_size = learning_rate * md 
        coefficient.append(m_curr)
        c_curr = c_curr - learning_rate * cd  # # step_size = learning_rate * cd 
        intercept.append(c_curr)
        # print('>epoch=%d, m=%.3f,c=%.3f, md=%.3f,cd=%.3f,step_size_m=%.3f,step_size_c=%.3f,cost=%.3f' % 
        #       (i,m_curr,c_curr, md,cd,md*learning_rate,cd*learning_rate,(cost)))
    return intercept,coefficient,cost_val

x = np.array([3,4,6,12,9,15,10,1,8,13])
y = np.array([16,29,43,65,51,89,57,9,53,68])
epochs = 100
n = len(x)
learning_rate = 0.001
data=Simple_LinearRegression_Gradient_Descent(x,y,epochs,learning_rate,n)
# 画出损失函数的曲线, x轴为coefficient, y轴为cost_val
data=pd.DataFrame(data).T
print(data)
plt.plot(data[1],data[2],label='beta-MSE')

 

# 使用sklearn进行线性拟合
# from sklearn.model_selection import train_test_split
lr = LinearRegression()
# x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2, random_state=42)
# x_train=x_train.reshape(-1, 1)
# y_train=y_train.reshape(-1, 1)
x=x.reshape(-1, 1)
lr.fit(x,y)


lr.intercept_,lr.coef_


# 对x进行预测
y_predict = lr.predict(x)
y_predict,y

 

# 使用sklearn计算mse
mse=mean_squared_error(y,y_predict)
mse

 

from sklearn.linear_model import ElasticNet
# 使用ElasticNet进行拟合和预测, 并计算mse
print(x,y)
model=ElasticNet()
model.fit(x,y)
print(model.intercept_,model.coef_)
y_predict = model.predict(x)
mse_=mean_squared_error(y,y_predict)
print(mse)