from random import random,randint
import math

# 根據(jù)等級(jí)和年代對(duì)價(jià)格進(jìn)行模擬
def wineprice(rating,age):
 peak_age=rating-50

 # 根據(jù)等級(jí)計(jì)算價(jià)格
 price=rating/2
 if age>peak_age:
  # 經(jīng)過(guò)“峰值年”，后續(xù)5年里其品質(zhì)將會(huì)變差
  price=price*(5-(age-peak_age)/2)
 else:
  # 價(jià)格在接近“峰值年”時(shí)會(huì)增加到原值的5倍
  price=price*(5*((age+1)/peak_age))
 if price<0: price=0
 return price

# 生成一批模式數(shù)據(jù)代表樣本數(shù)據(jù)集
def wineset1():
 rows=[]
 for i in range(300):
  # 隨機(jī)生成年代和等級(jí)
  rating=random()*50+50
  age=random()*50

  # 得到一個(gè)參考價(jià)格
  price=wineprice(rating,age)

  # 添加一些噪音
  price*=(random()*0.2+0.9)

  # 加入數(shù)據(jù)集
  rows.append({'input':(rating,age),'result':price})
 return rows

# 使用歐幾里得距離，定義距離
def euclidean(v1,v2):
 d=0.0
 for i in range(len(v1)):
  d+=(v1[i]-v2[i])**2
 return math.sqrt(d)

# 計(jì)算給預(yù)測(cè)商品和原數(shù)據(jù)集中任一其他商品間的距離。data原數(shù)據(jù)集，vec1預(yù)測(cè)商品
def getdistances(data,vec1):
 distancelist=[]

 # 遍歷數(shù)據(jù)集中的每一項(xiàng)
 for i in range(len(data)):
  vec2=data[i]['input']

  # 添加距離到距離列表
  distancelist.append((euclidean(vec1,vec2),i))

 # 距離排序
 distancelist.sort()
 return distancelist #返回距離列表

# 對(duì)距離值最小的前k個(gè)結(jié)果求平均
def knnestimate(data,vec1,k=5):
 # 得到經(jīng)過(guò)排序的距離值
 dlist=getdistances(data,vec1)
 avg=0.0

 # 對(duì)前k項(xiàng)結(jié)果求平均
 for i in range(k):
  idx=dlist[i][1]
  avg+=data[idx]['result']
 avg=avg/k
 return avg

def inverseweight(dist,num=1.0,const=0.1):
 return num/(dist+const)

def subtractweight(dist,const=1.0):
 if dist>const:
  return 0
 else:
  return const-dist

def gaussian(dist,sigma=5.0):
 return math.e**(-dist**2/(2*sigma**2))

# 對(duì)距離值最小的前k個(gè)結(jié)果求加權(quán)平均
def weightedknn(data,vec1,k=5,weightf=gaussian):
 # 得到距離值
 dlist=getdistances(data,vec1)
 avg=0.0
 totalweight=0.0

 # 得到加權(quán)平均
 for i in range(k):
  dist=dlist[i][0]
  idx=dlist[i][1]
  weight=weightf(dist)
  avg+=weight*data[idx]['result']
  totalweight+=weight
 if totalweight==0: return 0
 avg=avg/totalweight
 return avg

if __name__=='__main__': #只有在執(zhí)行當(dāng)前模塊時(shí)才會(huì)運(yùn)行此函數(shù)
 data = wineset1()  #創(chuàng)建第一批數(shù)據(jù)集
 result=knnestimate(data,(95.0,3.0)) #根據(jù)最近鄰直接求平均進(jìn)行預(yù)測(cè)
 print(result)

 result=weightedknn(data,(95.0,3.0),weightf=inverseweight) #使用反函數(shù)做權(quán)值分配方法，進(jìn)行加權(quán)平均
 print(result)
 result = weightedknn(data, (95.0, 3.0), weightf=subtractweight) # 使用減法函數(shù)做權(quán)值分配方法，進(jìn)行加權(quán)平均
 print(result)
 result = weightedknn(data, (95.0, 3.0), weightf=gaussian) # 使用高斯函數(shù)做權(quán)值分配方法，進(jìn)行加權(quán)平均
 print(result)

# 劃分?jǐn)?shù)據(jù)。test測(cè)試數(shù)據(jù)集占的比例。其他數(shù)據(jù)集為訓(xùn)練數(shù)據(jù)
def dividedata(data,test=0.05):
 trainset=[]
 testset=[]
 for row in data:
  if random()<test:
   testset.append(row)
  else:
   trainset.append(row)
 return trainset,testset

# 使用數(shù)據(jù)集對(duì)使用算法進(jìn)行預(yù)測(cè)的結(jié)果的誤差進(jìn)行統(tǒng)計(jì)，一次判斷算法好壞。algf為算法函數(shù)，trainset為訓(xùn)練數(shù)據(jù)集，testset為預(yù)測(cè)數(shù)據(jù)集
def testalgorithm(algf,trainset,testset):
 error=0.0
 for row in testset:
  guess=algf(trainset,row['input']) #這一步要和樣本數(shù)據(jù)的格式保持一致，列表內(nèi)個(gè)元素為一個(gè)字典，每個(gè)字典包含input和result兩個(gè)屬性。而且函數(shù)參數(shù)是列表和元組
  error+=(row['result']-guess)**2
  #print row['result'],guess
 #print error/len(testset)
 return error/len(testset)

# 將數(shù)據(jù)拆分和誤差統(tǒng)計(jì)合并在一起。對(duì)數(shù)據(jù)集進(jìn)行多次劃分，并驗(yàn)證算法好壞
def crossvalidate(algf,data,trials=100,test=0.1):
 error=0.0
 for i in range(trials):
  trainset,testset=dividedata(data,test)
  error+=testalgorithm(algf,trainset,testset)
 return error/trials

if __name__=='__main__': #只有在執(zhí)行當(dāng)前模塊時(shí)才會(huì)運(yùn)行此函數(shù)
 data = wineset1()  #創(chuàng)建第一批數(shù)據(jù)集
 print(data)
  error = crossvalidate(knnestimate,data) #對(duì)直接求均值算法進(jìn)行評(píng)估
 print('平均誤差：'+str(error))

 def knn3(d,v): return knnestimate(d,v,k=3) #定義一個(gè)函數(shù)指針。參數(shù)為d列表，v元組
 error = crossvalidate(knn3, data)   #對(duì)直接求均值算法進(jìn)行評(píng)估
 print('平均誤差：' + str(error))

 def knninverse(d,v): return weightedknn(d,v,weightf=inverseweight) #定義一個(gè)函數(shù)指針。參數(shù)為d列表，v元組
 error = crossvalidate(knninverse, data)   #對(duì)使用反函數(shù)做權(quán)值分配方法進(jìn)行評(píng)估
 print('平均誤差：' + str(error))

# 構(gòu)建新數(shù)據(jù)集，模擬不同類型變量的問(wèn)題
def wineset2():
 rows=[]
 for i in range(300):
  rating=random()*50+50 #酒的檔次
  age=random()*50   #酒的年限
  aisle=float(randint(1,20)) #酒的通道號(hào)（無(wú)關(guān)屬性）
  bottlesize=[375.0,750.0,1500.0][randint(0,2)] #酒的容量
  price=wineprice(rating,age) #酒的價(jià)格
  price*=(bottlesize/750)
  price*=(random()*0.2+0.9)
  rows.append({'input':(rating,age,aisle,bottlesize),'result':price})
 return rows

# 按比例對(duì)屬性進(jìn)行縮放，scale為各屬性的值的縮放比例。
def rescale(data,scale):
 scaleddata=[]
 for row in data:
  scaled=[scale[i]*row['input'][i] for i in range(len(scale))]
  scaleddata.append({'input':scaled,'result':row['result']})
 return scaleddata

# 生成成本函數(shù)。閉包
def createcostfunction(algf,data):
 def costf(scale):
  sdata=rescale(data,scale)
  return crossvalidate(algf,sdata,trials=10)
 return costf

weightdomain=[(0,10)]*4  #將縮放比例這個(gè)題解的取值范圍設(shè)置為0-10，可以自己設(shè)定，用于優(yōu)化算法

if __name__=='__main__': #只有在執(zhí)行當(dāng)前模塊時(shí)才會(huì)運(yùn)行此函數(shù)
 #========縮放比例優(yōu)化===
 data = wineset2() # 創(chuàng)建第2批數(shù)據(jù)集
 print(data)
 import optimization
 costf=createcostfunction(knnestimate,data)  #創(chuàng)建成本函數(shù)
 result = optimization.annealingoptimize(weightdomain,costf,step=2) #使用退火算法尋找最優(yōu)解
 print(result)

def wineset3():
 rows=wineset1()
 for row in rows:
  if random()<0.5:
   # 葡萄酒是從折扣店購(gòu)買的
   row['result']*=0.6
 return rows

# 計(jì)算概率。data樣本數(shù)據(jù)集，vec1預(yù)測(cè)數(shù)據(jù)，low，high結(jié)果范圍，weightf為根據(jù)距離進(jìn)行權(quán)值分配的函數(shù)
def probguess(data,vec1,low,high,k=5,weightf=gaussian):
 dlist=getdistances(data,vec1) #獲取距離列表
 nweight=0.0
 tweight=0.0

 for i in range(k):
  dist=dlist[i][0] #距離
  idx=dlist[i][1] #索引號(hào)
  weight=weightf(dist) #權(quán)值
  v=data[idx]['result'] #真實(shí)結(jié)果

  # 當(dāng)前數(shù)據(jù)點(diǎn)位于指定范圍內(nèi)么？
  if v>=low and v<=high:
   nweight+=weight #指定范圍的權(quán)值之和
  tweight+=weight  #總的權(quán)值之和
 if tweight==0: return 0

 # 概率等于位于指定范圍內(nèi)的權(quán)重值除以所有權(quán)重值
 return nweight/tweight

from pylab import *

# 繪制累積概率分布圖。data樣本數(shù)據(jù)集，vec1預(yù)測(cè)數(shù)據(jù)，high取值最高點(diǎn)，k近鄰范圍，weightf權(quán)值分配
def cumulativegraph(data,vec1,high,k=5,weightf=gaussian):
 t1=arange(0.0,high,0.1)
 cprob=array([probguess(data,vec1,0,v,k,weightf) for v in t1]) #預(yù)測(cè)產(chǎn)生的不同結(jié)果的概率
 plot(t1,cprob)
 show()

# 繪制概率密度圖
def probabilitygraph(data,vec1,high,k=5,weightf=gaussian,ss=5.0):
 # 建立一個(gè)代表價(jià)格的值域范圍
 t1=arange(0.0,high,0.1)

 # 得到整個(gè)值域范圍內(nèi)的所有概率
 probs=[probguess(data,vec1,v,v+0.1,k,weightf) for v in t1]

 # 通過(guò)加上近鄰概率的高斯計(jì)算結(jié)果，對(duì)概率值做平滑處理
 smoothed=[]
 for i in range(len(probs)):
  sv=0.0
  for j in range(0,len(probs)):
   dist=abs(i-j)*0.1
   weight=gaussian(dist,sigma=ss)
   sv+=weight*probs[j]
  smoothed.append(sv)
 smoothed=array(smoothed)

 plot(t1,smoothed)
 show()

if __name__=='__main__': #只有在執(zhí)行當(dāng)前模塊時(shí)才會(huì)運(yùn)行此函數(shù)

 data = wineset3() # 創(chuàng)建第3批數(shù)據(jù)集
 print(data)
 cumulativegraph(data,(1,1),120) #繪制累積概率密度
 probabilitygraph(data,(1,1),6) #繪制概率密度圖