机器学习基石 作业1 实现PLA和Pocket算法

#!/usr/bin/env python
# -*- coding: utf-8 -*-
"""
__title__ = 'PLA and Pocket Algorithm'
__author__ = 'w1d2s'
__mtime__ = '2015/10/5'
"""
import sys
import string
import random
import time
from numpy import *
from numpy.linalg import *

def Data_Pretreatment(path):
rawData = open(path).readlines()
dataNum = len(rawData)
dataDim = len(rawData[0].strip().split(' ')) + 1
dataIdx = 0
X = zeros([dataNum, dataDim])
X[:, 0] = 1
Y = zeros(dataNum)
print(dataNum, dataDim)
for line in rawData:
temp = line.strip().split('\t')
temp[0] = temp[0].split(' ')
Y[dataIdx] = string.atof(temp[1])
X[dataIdx, 1:] = double(temp[0])
dataIdx += 1
return (X, Y)

def PLA_Cycle(X, Y, eta, IsRandom):
# X : input set of training data, Y : output set of..., eta : learning ratio (0 ~ 1)
# IsRandom == False -> Naive Cycle,  IsRandom == True -> Random Cycle
(dataNum, dataDim) = X.shape
W = zeros(dataDim)
permutation = range(0, dataNum)
if IsRandom:
random.shuffle(permutation)
else:
pass
upDateTimes = 0
lastUpDateIdx = 0;
pmtIdx = 0
dataIdx = 0
iteCnt = 0
halt = False
while not halt:
dataIdx = permutation[pmtIdx]
dotProduct = dot(W, X[dataIdx])
if dotProduct * Y[dataIdx] > 0:
if dataIdx == lastUpDateIdx:
halt = True
else:
pass
else:
# PLA update: W(t+1) = W(t) + eta*Y(n)*X(n)
W += eta * Y[dataIdx] * X[dataIdx]
upDateTimes += 1
lastUpDateIdx = dataIdx
pmtIdx = (pmtIdx + 1) % dataNum
iteCnt += 1
# print(iteCnt, W)
print('upDateTimes: ', upDateTimes, '\n')
return (W, upDateTimes)

def Is_Same_Sign(W, vector, y):
if (dot(W, vector) > 0 and y > 0) or (dot(W, vector) <= 0 and y < 0):
return True
else:
return False

def Get_Err_Num(X, Y, W):
(dataNum, dataDim) = X.shape
Sum = 0
for i in range(0, dataNum):
if not Is_Same_Sign(W, X[i], Y[i]):
Sum += 1
return Sum

def Pocket_Algo(X, Y, eta, iterateTimes):
(dataNum, dataDim) = X.shape
W_p = zeros(dataDim)
id_p = 0
ErrNum_p = Get_Err_Num(X, Y, W_p)
W = zeros(dataDim)
ErrNum = 0
iterate = 0
dataIdx = 0
#random.seed(int(time.time() % 3000))
while iterate <= iterateTimes:
dataIdx = random.randint(0, dataNum - 1)
if not Is_Same_Sign(W, X[dataIdx], Y[dataIdx]):
iterate += 1
W = W + eta * Y[dataIdx] * X[dataIdx] #!!!! W += eta * Y[dataIdx] * X[dataIdx] will change value of W_p
ErrNum = Get_Err_Num(X, Y, W)
if ErrNum < ErrNum_p:
W_p = W
ErrNum_p = ErrNum
# print 'ErrNum_p: ' + str(ErrNum_p)
return W_p

if __name__ == '__main__':
"""
(X, Y) = Data_Pretreatment('train.txt')
#W = PLA_Cycle(X, Y, 1, 'N')
#print W
sum = 0
for i in range(0, 2000):
(W, upDateTimes) = PLA_Cycle(X, Y, 0.5, True)
sum += upDateTimes
print str(int(sum/2000))
"""
(X_train, Y_train) = Data_Pretreatment('18_train.txt')
(X_test, Y_test) = Data_Pretreatment('18_test.txt')
errSum = 0

for times in range(0, 100):
#random.seed(times)
W_pocket = Pocket_Algo(X_train, Y_train, 1, 100)
errNum_out = Get_Err_Num(X_test, Y_test, W_pocket)
errSum += errNum_out
print str(errNum_out)

errRate_out = errSum/100
print str(errRate_out)