机器学习基石 PLA 作业1

主要参考：
https://blog.csdn.net/sjz_hahalala479/article/details/81003517
https://blog.csdn.net/qq_25037903/article/details/83861118
小菜鸡开始学习ML了。。。
15.

import numpy as np
def getDataset(l,s):
    X = np.zeros((l,5)) #多一个x0
    Y = np.zeros((l,1))
    for i in range(l):
        ss = s[i].strip().split()
        #print (type(ss[0]))
        X[i][0] = 1
        X[i][1] = ss[0]
        X[i][2] = ss[1]
        X[i][3] = ss[2]
        X[i][4] = ss[3]
        Y[i][0] = ss[4]
    return X,Y
def sign(X,W):
    w = np.matrix(W)
    x = np.matrix(X)
    if np.dot(W,X) >0:
        #print (W)
        return 1
    else:
        return -1
def pla(l,s,W):
    X,Y = getDataset(l,s)
    iteration = 1
    updates = 0
    while(iteration!=0):
        iteration = 0
        for i in range(l):
            if sign(X[i],W) != Y[i]:
                W = W+Y[i]*X[i]
                iteration +=1
                #break #这里不是break 之前搞错了算法流程 一直不对 是所有点顺序遍历后重新迭代
        updates += iteration
    return updates

with open('result.txt','r') as f:
    s = f.readlines()
l = len(s) #行
W = np.zeros(5)
print(pla(l,s,W))

16、17

import numpy as np
import random
def getDataset(l,s):
    X = np.zeros((l,5)) #多一个x0
    Y = np.zeros((l,1))
    for i in range(l):
        ss = s[i].strip().split()
        #print (type(ss[0]))
        X[i][0] = 1
        X[i][1] = ss[0]
        X[i][2] = ss[1]
        X[i][3] = ss[2]
        X[i][4] = ss[3]
        Y[i][0] = ss[4]
    return X,Y
def sign(X,W):
    w = np.matrix(W)
    x = np.matrix(X)
    if np.dot(W,X) >0:
        #print (W)
        return 1
    else:
        return -1
def pla(l,s,W,randomlist):
    X,Y = getDataset(l,s)
    iteration = 1
    updates = 0
    while(iteration!=0):
        iteration = 0
        random.shuffle(randomlist)
        for i in randomlist:
            if sign(X[i],W) != Y[i]:
                W = W+Y[i]*X[i]*0.25
                iteration +=1
                #break
        updates += iteration
    return updates

if __name__=='__main__':
    with open('result.txt','r') as f:
        s = f.readlines()
    l = len(s) #行数
    m=[]
    rl = list(range(l))
    for i in range(2000):
   		W = np.zeros(5)
        m.append(pla(l,s,W,rl))
    mm = np.array(m)
    print(np.mean(mm))

18、19、20

import numpy as np
import random
def getDataset(l,s):
    X = np.zeros((l,5)) #多一个x0=1
    Y = np.zeros((l,1))
    for i in range(l):
        ss = s[i].strip().split()
        #print (type(ss[0]))
        X[i][0] = 1
        X[i][1] = ss[0]
        X[i][2] = ss[1]
        X[i][3] = ss[2]
        X[i][4] = ss[3]
        Y[i][0] = ss[4]
    return X,Y
def sign(X,W):
    if np.dot(W,X) >0:
        return 1
    else:
        return -1
def mistake(W,X,Y):
    cnt = 0
    ls= len(X)
    for i in range(ls):
        res = sign(X[i],W)
        if res!=Y[i]:
            cnt +=1
    return cnt/ls
def pla(l,s,W,randomlist):
    X,Y = getDataset(l,s)
    Wt = W
    updates = 100
    while(updates>0):
        iteration = 0
        random.shuffle(randomlist)
        for i in randomlist:
            if sign(X[i],Wt) != Y[i]:  #这里比较的是Wt 就是一直更新的线 而不是最好的线 最好的W是在不断更新中保存的 
                Wt = Wt+Y[i]*X[i]
                #print(Wt,W)
                miscnt1= mistake(Wt,X,Y)  #现有线向量的错误率
                miscnt2 = mistake(W,X,Y)  #原有线向量的错误率
                updates -= 1
                if miscnt1<miscnt2:
                    W = Wt
            if updates==0:
                break
        if updates ==0:
            break
    return W

if __name__=='__main__':
    with open('result.txt','r') as f:
        s = f.readlines()
    with open('mytest.txt','r') as ff:
        ss =ff.readlines()
    l = len(s) #行数
    ll = len(ss)
    tstX, tstY = getDataset(ll, ss)
    m =[]
    rl = list(range(l)) #l索引list
    for i in range(20):
        W = np.zeros(5)
        Wm = pla(l,s,W,rl)
        m.append(mistake(Wm,tstX,tstY))
    a = np.array(m)
    print(a.mean())

感悟：debug de的我已经是个dead person了