吴恩达ex1——单变量线性回归

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

#加载数据
df=pd.read_csv('ex1data1-Copy1.txt',header=None,names=['population','profit'])
df.head()

	population	profit
0	6.1101	17.5920
1	5.5277	9.1302
2	8.5186	13.6620
3	7.0032	11.8540
4	5.8598	6.8233

df_np=df.values
df_np=np.insert(df_np,0,1.0,axis=1)

df_np.shape

(97, 3)

X=df_np[:,0:2]
X=np.matrix(X)
X.shape

(97, 2)

#绘图
x1=df.iloc[:,0].values
y=df.iloc[:,1].values
plt.scatter(x1,y,color='r')
plt.xlabel('population')
plt.ylabel('profit')
plt.show()

#将y变成列向量
y=y.reshape(97,1)
y=np.matrix(y)
y.shape

(97, 1)

梯度下降实现

'X:97x2'
'theta：列向量,2x1 '

#定义损失函数
def cost(X,y,theta):
    inn=np.sum(np.power((X*theta-y),2))
    return inn/(2*len(y))
    

theta=[0.1,0.1]
theta=np.matrix(theta).reshape(2,1)
theta.shape

(2, 1)

loss_0=cost(X,y,theta)
loss_0

25.449553111855668

#更新theta
def GredientDec(X,y,theta,iters,alpha):
    parameters=X.shape[1]
    loss=np.zeros((iters,1))
    theta_fig=theta
    
    for a in range(iters):
        error=(X*theta-y)
        
        for j in range(parameters):
            term=np.sum(np.multiply(error,X[:,j]))
            theta[j]=theta[j]-(alpha*term)/len(y)
            loss[a]=cost(X,y,theta)
        
      
            
    return theta,loss

np.seterr(invalid='ignore')
theta,loss=GredientDec(X,y,theta=theta,iters=1000,alpha=0.01)
theta

matrix([[-3.78565572],
        [ 1.18197038]])

loss_new=cost(X,y,theta)

loss_new

4.478075461131649

x=np.linspace(X[:,1].min(),X[:,1].max(),100)
y_fig=theta[0]+theta[1]*x
y_fig=y_fig.reshape(100,1)

x1=df.iloc[:,0].values
y=df.iloc[:,1].values
plt.scatter(x1,y,color='r')
plt.plot(x,y_fig,color='k')
plt.xlabel('population')
plt.ylabel('profit')
plt.show()

绘制损失函数

iters=1000
plt.plot(np.arange(iters),loss,color='r')
plt.xlabel('numbers of iter')
plt.ylabel('loss of J(θ)')
plt.show()

绘制等高线图

J=[]
for i in np.arange(-10,10,0.1):
    for j in np.arange(-10,10,0.1):
        theta=np.matrix([i,j]).reshape(2,1)
        J.append(cost(X,y,theta=theta))

J=np.array(J).reshape(200,200)

plt.contour(np.arange(-10,10,0.1),np.arange(-10,10,0.1),J,levels=20)

<matplotlib.contour.QuadContourSet at 0x173d9b6a8b0>