机器学习基石 作业1 程序题（15-20）

import numpy as np
import random
def read_data(path):
    f1=open(path)
    x_matrix=[]
    y_matrix=[]
    for i in f1:
        x=[1]
        for j in i.split('\t')[0].split():
            x.append(float(j))
        y_matrix.append(int(i.strip().split('\t')[1]))
        x_matrix.append(x)
    x_matrix=np.array(x_matrix)
    #print(x_matrix)
    return x_matrix,y_matrix
def sign(x,w):
    if np.dot(x, w)[0]<=0:
        return -1
    else:
        return 1
def naive_PLA(x_matrix,y_matrix):
    sum=len(x_matrix)
    length=len(x_matrix[0])
    w=np.zeros((length,1))
    #print(w)
    count=0
    s=0
    flag=0
    while True:
        for i in range(sum):
            s+=1
            #print(np.dot(x_matrix[i], w)[0]*y_matrix[i])
            if sign(x_matrix[i], w)!=y_matrix[i]:
                #print(w,x_matrix[i],y_matrix[i])
                w+=np.matrix(x_matrix[i]).T*y_matrix[i]
                count+=1
                s=0
            if s==sum:
                flag=1
                break
        if flag==1:
            break
    return count
def random_PLA(x_matrix,y_matrix):
    sum=len(x_matrix)
    length=len(x_matrix[0])
    w=np.zeros((length,1))
    order=range(sum)
    #print(order)
    random_seed=random.sample(order,sum)
    #print(random_seed)
    count=0
    s=0
    flag=0
    while True:
        for i in random_seed:
            s+=1
            #print(np.dot(x_matrix[i], w)[0]*y_matrix[i])
            if sign(x_matrix[i], w)!=y_matrix[i]:
                #print(w,x_matrix[i],y_matrix[i])
                w+=np.matrix(x_matrix[i]).T*y_matrix[i]
                count+=1
                s=0
            if s==sum:
                flag=1
                break
        if flag==1:
            break
    return count

def weighted_random_PLA(x_matrix,y_matrix,eta):
    sum=len(x_matrix)
    length=len(x_matrix[0])
    w=np.zeros((length,1))
    order=range(sum)
    #print(order)
    random_seed=random.sample(order,sum)
    #print(random_seed)
    count=0
    s=0
    flag=0
    while True:
        for i in random_seed:
            s+=1
            #print(np.dot(x_matrix[i], w)[0]*y_matrix[i])
            if sign(x_matrix[i], w)!=y_matrix[i]:
                #print(w,x_matrix[i],y_matrix[i])
                w+=np.matrix(x_matrix[i]).T*y_matrix[i]*eta
                count+=1
                s=0
            if s==sum:
                flag=1
                break
        if flag==1:
            break
    return count

if __name__=='__main__':

    x_matrix, y_matrix=read_data('ntumlone_hw1_hw1_15_train.dat')

    #count=naive_PLA(x_matrix,y_matrix)       #Question 15
    #print(count)

    # sum=0                                     #Question 16
    # for i in range(2000):
    #     sum+=random_PLA(x_matrix,y_matrix)
    #     print()
    # print(sum/2000)

    sum=0                                     #Question 17
    for i in range(2000):
        sum+=weighted_random_PLA(x_matrix,y_matrix,0.5)
    print(sum/2000)

import numpy as np
import random
import copy
def read_data(path):
    f1=open(path)
    x_matrix=[]
    y_matrix=[]
    for i in f1:
        x=[1]
        for j in i.split('\t')[0].split():
            x.append(float(j))
        y_matrix.append(int(i.strip().split('\t')[1]))
        x_matrix.append(x)
    x_matrix=np.array(x_matrix)
    #print(x_matrix)
    return x_matrix,y_matrix
def sign(x,w):
    if np.dot(x, w)[0]<=0:
        return -1
    else:
        return 1
def test(w,x_matrix,y_matrix,sum):
    count=0
    for i in range(sum):
        if sign(x_matrix[i],w) !=y_matrix[i]:
            count+=1
    return count
def random_pocket(x_matrix,y_matrix,updates):
    sum=len(x_matrix)
    length=len(x_matrix[0])
    order=range(sum)
    random_seed=random.sample(order,sum)
    bestW=np.zeros((length,1))
    w= np.zeros((length, 1))
    bestCount=501
    update=0

    while update<updates:
        for i in random_seed:
            if sign(x_matrix[i], w)!=y_matrix[i]:
                update += 1
                w=w+np.matrix(x_matrix[i]).T*y_matrix[i]
                count=test(w,x_matrix,y_matrix,sum)
                if count<bestCount:
                    bestCount=count
                    bestW=w
            if update==updates:
                break
    return bestW

def random_PLA(x_matrix,y_matrix):
    sum=len(x_matrix)
    length=len(x_matrix[0])
    w=np.zeros((length,1))
    order=range(sum)
    #print(order)
    random_seed=random.sample(order,sum)
    #print(random_seed)
    count=0
    while True:
        for i in random_seed:
            if sign(x_matrix[i], w)!=y_matrix[i]:
                w+=np.matrix(x_matrix[i]).T*y_matrix[i]
                count+=1
            if count==50:
                break
        if count == 50:
            break
    return w

if __name__=='__main__':

    x_matrix, y_matrix=read_data('train.txt')
    x_test,y_test=read_data('test.txt')

    sum=len(x_test)
    error=0

    # for i in range(2000):                        #Question 18
    #     print(i)
    #     w=random_pocket(x_matrix, y_matrix, 50)
    #     count=test(w,x_test,y_test,sum)
    #     error+=count
    # print(float(error)/sum/2000)

    # for i in range(2000):                        #Question 19
    #     print(i)
    #     w=random_PLA(x_matrix, y_matrix)
    #     count=test(w,x_test,y_test,sum)
    #     error+=count
    # print(float(error)/sum/2000)

    for i in range(2000):                          #Question 20
        print(i)
        w=random_pocket(x_matrix, y_matrix, 100)
        count=test(w,x_test,y_test,sum)
        error+=count
    print(float(error)/sum/2000)