# KNN算法思路：
#-----------------------------------------------------#
#step1:读入数据，存储为链表
#step2:数据预处理，包括缺失值处理、归一化等
#step3:设置K值
#step4:计算待测样本与所有样本的距离(二值、序数、连续)
#step5:投票决定待测样本的类别
#step6:利用测试集测试正确率
#-----------------------------------------------------#
注：因为是python的初学者，可能很多高级的用法还不会，所以把python代码写的像C还请大家不要吐槽。同时希望大家指出其中的错误和有待提高的地方，大家一起进步才是最棒的。
说明：数据集采自著名UCI数据集库 http://archive.ics.uci.edu/ml/datasets/Adult
# Author :CWX
# Date :2015/9/1
# Function: A classifier which using KNN algorithm
import math
attributes = {"age":0,"workclass":1,"fnlwg":2,"education":3,"education-num":4,
"marital-status":5,"occupation":6,"relationship":7,"race":8,
"sex":9,"capital-gain":10,"capital-loss":11,"hours-per-week":12,
"native-country":13,"salary":14
}
def read_txt(filename):
#read data and convert it into list
items = []
fp = open(filename,‘r‘)
lines = fp.readlines()
for line in lines:
line = line.strip(‘\n‘)
items.append(line)
fp.close()
i = 0
b = []
for i in range(len(items)):
b.append(items[i].split(‘,‘))
return b
def computeNa(items):
# detect missing value in list and handle it
# items - an whole list
for item in items[:]:
if item.count(‘ ?‘) > 0:
items.remove(item)
# if item.count(‘ ?‘) >= -1:
# items.remove(item)
return items
def disCal(lst1,lst2,type):
# calculting distance between lst1 and lst2
distance = 0;
if type == "Manhattan" or type == "manhattan":
for i in range(len(lst2) - 1):
distance += abs(lst1[i] - lst2[i])
elif type == "Elucildean" or type == "elucildean":
for i in range(len(lst2) - 1):
distance += math.sqrt((lst1[i] - lst2[i])**2)
else:
print "Error in type name"
distance = -1
return distance
def computeContinous(datalist,attribute):
# compute continous attributes in list
min_val = int(datalist[0][attribute])
max_val = int(datalist[0][attribute])
for items in datalist:
if int(items[attribute]) < min_val:
min_val = int(items[attribute])
elif int(items[attribute]) > max_val:
max_val = int(items[attribute])
for items in datalist[:]:
items[attribute] = (int(items[attribute]) - min_val) / float(max_val - min_val)
return datalist
def computeOrdinal(datalist,attribute,level):
# compute ordinal attribute in datalist
level_dict = {}
for i in range(len(level)):
level_dict[level[i]] = float(i) / (len(level) - 1)
# level_dict[level[i]] = i
for items in datalist[:]:
items[attribute] = level_dict[items[attribute]]
return datalist
def KnnAlgorithm(dataTrain,sample,attribute,k):
mergeData = dataTrain
mergeData.append(sample)
data = preProcessing(mergeData)
distance = []
for i in range(len(data)-2):
distance.append(disCal(data[i],data[len(data)-1],"Elucildean"))
copy_dis = distance[:] # notice : not copy_dis = distance ,if it will be wrong
distance.sort()
class_dict = {"Yes":0,"No":0}
for i in range(k):
index = copy_dis.index(distance[i])
if data[index][attribute] == " >50K":
class_dict["Yes"] += 1
else:
class_dict["No"] += 1
if class_dict["Yes"] > class_dict["No"]:
print "sample‘s salary >50K"
else:
print "sample‘s salary <=50K"
def preProcessing(datalist):
b = computeNa(datalist)
b = computeContinous(b,attributes["age"])
workclass_level = [" Private"," Self-emp-not-inc"," Self-emp-inc"," Federal-gov"," Local-gov"," State-gov"," Without-pay"," Never-worked"]
b = computeOrdinal(b,attributes["workclass"],workclass_level)
b = computeContinous(b,attributes["fnlwg"])
education_level =[" Bachelors"," Some-college"," 11th"," HS-grad"," Prof-school",
" Assoc-acdm"," Assoc-voc"," 9th"," 7th-8th"," 12th"," Masters"," 1st-4th"," 10th"," Doctorate"," 5th-6th"," Preschool"]
b = computeOrdinal(b,attributes["education"],education_level)
b = computeContinous(b,attributes["education-num"])
marital_status_level = [" Married-civ-spouse"," Divorced"," Never-married"," Separated"," Widowed"," Married-spouse-absent"," Married-AF-spouse"]
b = computeOrdinal(b,attributes["marital-status"],marital_status_level)
occupation_level = [" Tech-support"," Craft-repair"," Other-service"," Sales"," Exec-managerial"," Prof-specialty"," Handlers-cleaners",
" Machine-op-inspct"," Adm-clerical"," Farming-fishing"," Transport-moving"," Priv-house-serv"," Protective-serv"," Armed-Forces"]
b = computeOrdinal(b,attributes["occupation"],occupation_level)
relationship_level = [" Wife"," Own-child"," Husband"," Not-in-family"," Other-relative"," Unmarried"]
b = computeOrdinal(b,attributes["relationship"],relationship_level)
race_level = [" White"," Asian-Pac-Islander"," Amer-Indian-Eskimo"," Other"," Black"]
b = computeOrdinal(b,attributes["race"],race_level)
sex_level = [" Female", " Male"]
b = computeOrdinal(b,attributes["sex"],sex_level)
b = computeContinous(b,attributes["capital-gain"])
b = computeContinous(b,attributes["capital-loss"])
b = computeContinous(b,attributes["hours-per-week"])
native_country_level = [" United-States"," Cambodia"," England"," Puerto-Rico"," Canada"," Germany"," Outlying-US(Guam-USVI-etc)"," India",
" Japan"," Greece"," South"," China"," Cuba"," Iran"," Honduras"," Philippines"," Italy"," Poland"," Jamaica"," Vietnam"," Mexico"," Portugal",
" Ireland"," France"," Dominican-Republic"," Laos"," Ecuador"," Taiwan"," Haiti"," Columbia"," Hungary"," Guatemala"," Nicaragua"," Scotland",
" Thailand"," Yugoslavia"," El-Salvador"," Trinadad&Tobago"," Peru"," Hong"," Holand-Netherlands"]
b = computeOrdinal(b,attributes["native-country"],native_country_level)
return b
def assessment(dataTrain,dataTest,atrribute,k):
mergeData = computeNa(dataTrain)
len_train = len(mergeData)
mergeData.extend(computeNa(dataTest))
data = preProcessing(mergeData)
len_test = len(data) - len_train
res_dict = {"correct":0,"wrong":0}
for i in range(len_test):
distance = []
class_dict = {"Yes":0,"No":0}
for j in range(len_train):
distance.append(disCal(data[j],data[i+len_train],"Elucildean"))
copy_dis = distance[:]
distance.sort()
for m in range(k):
index = copy_dis.index(distance[m])
if data[index][atrribute] == " >50K":
class_dict["Yes"] += 1
else:
class_dict["No"] += 1
if class_dict["Yes"] > class_dict["No"] and mergeData[i+len_train][atrribute] == " >50K.": #Attention : in train data in the end of lines there is a "."
res_dict["correct"] += 1
elif mergeData[i+len_train][atrribute] == " <=50K." and class_dict["Yes"] < class_dict["No"]:
res_dict["correct"] += 1
else:
res_dict["wrong"] += 1
correct_ratio = float(res_dict["correct"]) / (res_dict["correct"] + res_dict["wrong"])
print "correct_ratio = ",correct_ratio
filename = "H:\BaiduYunDownload\AdultDatasets\Adult_data.txt"
#sample = [" 80"," Private"," 226802"," 11th"," 7"," Never-married"," Machine-op-inspct"," Own-child"," Black"," Male"," 0"," 0"," 40"," United-States"," <=50K"]
sample = [" 65"," Private"," 184454"," HS-grad"," 9"," Married-civ-spouse"," Machine-op-inspct"," Husband"," White"," Male"," 6418"," 0"," 40"," United-States"," >50K"]
# this samples salary <=50K#
# filename = "D:\MyDesktop-HnH\data.txt"
a = read_txt(filename)
print len(a)
k = 3
#KnnAlgorithm(a,sample,attributes["salary"],k)
trainName = "H:\BaiduYunDownload\AdultDatasets\Adult_test.txt"
trainData = read_txt(trainName)
#preProcessing(trainData)
assessment(a,trainData,attributes["salary"],k)结果：正确率 0.812416998672
运行时间 1小时20分钟
本文出自 “路遥” 博客，请务必保留此出处http://cwxfly.blog.51cto.com/6113982/1691241
原文地址：http://cwxfly.blog.51cto.com/6113982/1691241
