吴恩达 机器学习 ex1练习 python

1 输出一个5*5的单位矩阵

import numpy as np
a=np.eye(5)
print(a)

结果：

[[1. 0. 0. 0. 0.]
 [0. 1. 0. 0. 0.]
 [0. 0. 1. 0. 0.]
 [0. 0. 0. 1. 0.]
 [0. 0. 0. 0. 1.]]

2 单变量的线性回归

整个2的部分需要根据城市人口数量，预测开小吃店的利润
数据在ex1data1.txt里，第一列是城市人口数量，第二列是该城市小吃店利润。

2.1画出数据

import matplotlib.pyplot as plt
ex=open("E:\exer\python\wex1\ex1data1.txt","r")
ex1=ex.readlines()
x=[]
y=[]
for i in ex1:
    i=i.strip().split(",")
    x.append(float(i[0]))
    y.append(float(i[1]))
plt.scatter(x,y,s=5,c='#FF69B4',alpha=0.8)
plt.xlabel("population")
plt.ylabel("profit")
plt.savefig("ex1.jpg")
plt.show()
ex.close()

在练习这一部分时，遇到的问题是，数据进入数组后无法画出来，找了很多参考都没法解决，然后意识到在读取数据时，存储的是字符型数据，需要转换成浮点数。
此外，在保存图片时，savefig要写在show前面，否则打开的图片会显示出一片空白

结果图：

2.2梯度下降

import matplotlib.pyplot as plt
import random
ex=open("E:\exer\python\wex1\ex1data1.txt","r")
ex1=ex.readlines()
x=[]
y=[]
for i in ex1:
    i=i.strip().split(",")
    x.append(float(i[0]))
    y.append(float(i[1]))
ex.close()
random.seed(1000)
a=random.random()
b=random.random()
m=0
q=0.01
iters=[]
Jp=[]
h_x1=[]
while(m<=2000):
    J0 = 0
    J1 = 0
    J2 = 0
    for i in range(0,len(x)):
        h_x=a*x[i]+b
        J0=J0+(h_x-y[i])**2
        J1=(h_x-y[i])*x[i]+J1
        J2=J2+(h_x-y[i])
    J=J0/(2*len(x))
    m=m+1
    a=a-q*J1/len(x)
    b=b-q*J2/len(x)
    iters.append(m)
    Jp.append(J)
for i in range(0,len(x)):
    h_x1.append(a*x[i]+b)
plt.subplot(1,2,1)
plt.plot(iters,Jp)
plt.xlabel("iters")
plt.ylabel("lost")
plt.subplot(1,2,2)
plt.scatter(x,y)
plt.plot(x,h_x1,c='#FF69B4')
plt.xlabel("x")
plt.ylabel("y")
plt.savefig("ex1")
plt.show()
print(a,b)

结果图：

review:

import matplotlib.pyplot as plt
import random
import numpy as np
#read data
doc=open("E:\exer\python\exercise1\ex1data1.txt","r")
data=doc.readlines()
population=[]
profit=[]
for i in data:
    i=i.strip().split(",")
    population.append(float(i[0]))
    profit.append(float(i[1]))
doc.close()
m=len(population)
#scatter data
'''plt.scatter(population,profit)
plt.xlabel("population")
plt.ylabel("profit")
plt.show()'''
#initiate the parameter
random.seed(100)
a=random.random()
b=random.random()
#forward
l=0.01#步长
iteration=1000
J=[]
for i in range(iteration):
    y=np.multiply(a,population)+b
    J1=np.dot((y-profit),(y-profit))/(2*m)
    J.append(J1)
    a=a-l*np.dot((y-profit),population)/m
    b=b-l*sum((y-profit))/m
plt.subplot(1,2,1)
#plot loss function
plt.plot(range(0,iteration),J)
#plot the result
plt.subplot(1,2,2)
plt.scatter(population,profit,c="#FF69B4")
x=range(0,int(max(population)))
result=np.multiply(a,x)+b
plt.plot(x,result)
plt.show()

3.多元线性回归

iimport numpy as np
import random
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
# read data
alpha=0.1
exer=open("E:\exer\python\wex1\ex1data2.txt","r")
exer1=exer.readlines()
lens=len(exer1)
feature=np.zeros((lens,3))
price=np.zeros((lens,1))
i=0
random.seed(100)
a=random.random()
b=random.random()
c=random.random()
for j in exer1:
    j=j.strip().split(",")
    feature[i:]=[1,int(j[0]),int(j[1])]
    price[i]=int(j[2])
    i+=1
exer.close()
# data normalization
feature1=feature
for i in range(1,3):
    feature1[:,i]=(feature1[:,i]-np.mean(feature1[:,i]))/np.std(feature1[:,i])
price1=(price-np.mean(price))/np.std(price)
q=np.mat([a,b,c])
q=q.T
Q=[]
I=[]
i=0
while(i<=80):
    L1=np.zeros((1,3))
    J=0
    h = np.dot(feature1,q)
    for j in range(0,lens):
        J=float((h[j]-price1[j])**2/(2*lens))+J
        L1=L1+(float(h[j]-price1[j]))*feature1[j,:]
    L1=L1.T
    q=q-alpha*L1/lens
    Q.append(J)
    I.append(i)
    i+=1
h = feature1*q
fig1 = plt.figure()
ax1 = Axes3D(fig1)
ax1.scatter((feature1.T)[1,:], (feature1.T)[2,:], price1, c='r')
xx = np.arange(-2,2,0.1)
yy = np.arange(-2,2,0.1)
X, Y = np.meshgrid(xx, yy)
Z =a+b*X+c*Y
ax1.plot_surface(X, Y, Z, rstride=1, cstride=1,)
ax1.set_xlabel("house_size")
ax1.set_ylabel("num of bedrooms")
ax1.set_zlabel("price")
plt.show()
plt.plot(I,Q)
plt.xlabel("iteration")
plt.ylabel("J")
plt.show()