import pandas as pd
import numpy as np
import math 
import os
import time
import datetime

os.chdir(r'f:\zxx\pthon_work\CF')

def loadData():
 #讀入movies.dat, rating.dat,tags.dat
 #mnames=['movie_id','title','genres']
 #movies=pd.read_table(r'.\data\movies.dat',sep='::',header=None,names=mnames)

 rnames=['UserID','MovieID','Rating','Timestamp']
 all_ratings=pd.read_table(r'.\data\ratings.dat',sep='::',header=None,names=rnames,nrows=300000)

 #tnames=['UserID','MovieID','Tag','Timestamp']
 #tags=pd.read_table(r'.\data\tags.dat',sep='::',header=None,names=tnames)
 return all_ratings

#數(shù)據(jù)探索：rating
def data_alay(ratings):
 """rating nums10000054, 3, 
 示例 ： 1  122  5 838985046
 col:'UserID','MovieID','Rating','Timestamp'
  """
 #一個(gè)用戶只對(duì)一個(gè)電影打分一次
 UR=ratings.groupby([ratings['UserID'],ratings['MovieID']])
 len(UR.size)

#計(jì)算每部電影的平均打分,電影數(shù)10677
def avgRating(ratings):
 movies_mean=ratings['Rating'].groupby(ratings['MovieID']).mean()#計(jì)算所有用戶對(duì)電影X的平均打分
 movies_id=movies_mean.index
 movies_avg_rating=movies_mean.values
 return movies_id,movies_avg_rating,movies_mean

#計(jì)算電影相似度矩陣相，即建立10677*10677矩陣
def calculatePC(ratings):
 movies_id,movies_avg_rating,movies_mean=avgRating(ratings)
 #pc_mat=np.eye(3)#建立電影相似度單位矩陣
 pc_dic={}
 top_movie=len(movies_id)
 for i in range(0,top_movie):
  for j in range(i+1,top_movie):
   movieAID=movies_id[i]
   movieBID=movies_id[j]
   see_moviesA_user=ratings['UserID'][ratings['MovieID']==movieAID]
   see_moviesB_user=ratings['UserID'][ratings['MovieID']==movieBID]
   join_user=np.intersect1d(see_moviesA_user.values,see_moviesB_user.values)#同時(shí)給電影A、B評(píng)分的用戶
   movieA_avg=movies_mean[movieAID]
   movieB_avg=movies_mean[movieBID]
   key1=str(movieAID)+':'+str(movieBID)
   key2=str(movieBID)+':'+str(movieAID)
   value=twoMoviesPC(join_user,movieAID,movieBID,movieA_avg,movieB_avg,ratings)
   pc_dic[key1]=value   
   pc_dic[key2]=value      
   #pc_mat[i][i+1]=twoMoviesPC(join_user,movieAID,movieBID,movieA_avg,movieB_avg,ratings)
   #print ('---the %s, %d,%d:--movie %s--%s--pc is %f' % (key1,movieAID,movieBID,movieAID,movieBID,pc_dic[key1]))
 return pc_dic

#計(jì)算電影A與電影B的相似度，皮爾森相似度=sum(A-A^)*sum(B-B^)/sqrt(sum[(A-A^)*(A-A^)]*sum[(B-B^)*(B-B^)])
def twoMoviesPC(join_user,movieAID,movieBID,movieA_avg,movieB_avg,ratings):
 cent_AB_sum=0.0#相似度分子
 centA_sum=0.0#分母
 centB_sum=0.0#分母
 movieAB_pc=0.0#電影A,B的相似度
 count=0
 for u in range(len(join_user)):
  #print '---------',u
  count=count+1
  ratA=ratings['Rating'][ratings['UserID']==join_user[u]][ratings['MovieID']==movieAID].values[0]#用戶給電影A評(píng)分
  ratB=ratings['Rating'][ratings['UserID']==join_user[u]][ratings['MovieID']==movieBID].values[0]#用戶給電影B評(píng)分
  cent_AB=(ratA-movieA_avg)*(ratB-movieB_avg) #去均值中心化
  centA_square=(ratA-movieA_avg)*(ratA-movieA_avg) #去均值平方
  centB_square=(ratB-movieB_avg)*(ratB-movieB_avg)#去均值平方
  cent_AB_sum=cent_AB_sum+cent_AB
  centA_sum=centA_sum+centA_square
  centB_sum=centB_sum+centB_square
 if(centA_sum>0 and centB_sum>0 ):
  movieAB_pc=cent_AB_sum/math.sqrt(centA_sum*centB_sum)
 return movieAB_pc

"""
預(yù)測(cè)用戶U對(duì)那些電影感興趣。分三步，
 1）用戶U過去X天看過的電影。
 2）提出用戶U已看過的電影，根據(jù)用戶U過去看過的電影，計(jì)算用戶U對(duì)其他電影的打分.
 3) 拉去打分最高的的電影給用戶推薦。
預(yù)測(cè)用戶U對(duì)電影C的打分。分三步：（先只做這個(gè)）
 1）用戶U過去X天看過的電影。
 2）利用加權(quán)去中心化公式預(yù)測(cè)用戶U對(duì)電影C的打分.

"""
#日期處理： -3天，然后轉(zhuǎn)換為uinxtime
def timePro(last_rat_time,UserU):
 lastDate= datetime.datetime.fromtimestamp(last_rat_time[UserU]) #unix轉(zhuǎn)為日期
 date_sub3=lastDate+datetime.timedelta(days=-3)#減去3天
 unix_sub3=time.mktime(date_sub3.timetuple())#日期轉(zhuǎn)為unix
 return unix_sub3

#取用戶最后一次評(píng)分前3天評(píng)估的電影進(jìn)行預(yù)測(cè)
def getHisRat(ratings,last_rat_time,UserUID):
 unix_sub3= timePro(last_rat_time,UserUID)
 UserU_info=ratings[ratings['UserID']==UserUID][ratings['Timestamp']>unix_sub3]
 return UserU_info

#預(yù)測(cè)用戶U對(duì)電影C的打分
def hadSeenMovieByUser(UserUID,MovieA,ratings,pc_dic,movies_mean):
 pre_rating=0.0 
 last_rat_time=ratings['Timestamp'].groupby([ratings['UserID']]).max()#獲取用戶U最近一次評(píng)分日期
 UserU_info= getHisRat(ratings,last_rat_time,UserUID)#獲取用戶U過去看過的電影

 flag=0#表示新電影，用戶U是否給電影A打過分
 wmv=0.0#相似度*mv平均打分去均值后之和
 w=0.0#相似度之和
 movie_userU=UserU_info['MovieID'].values#當(dāng)前用戶看過的電影
 if MovieA in movie_userU:
  flag=1
  pre_rating=UserU_info['Rating'][UserU_info['MovieID']==MovieA].values
 else:
  for mv in movie_userU:
   key=str(mv)+':'+str(MovieA)
   rat_U_mv=UserU_info['Rating'][UserU_info['MovieID']==mv][UserU_info['UserID']==UserUID].values#用戶U對(duì)看過電影mv的打分
   wmv=(wmv+pc_dic[key]*(rat_U_mv-movies_mean[mv]))#相似度*mv平均打分去均值后之和
   w=(w+pc_dic[key])#看過電影與新電影相似度之和
   #print ('---have seen mv %d with new mv %d,%f,%f'%(mv,MovieA,wmv,w))   
  pre_rating=(movies_mean[MovieA]+wmv/w)
 print ('-flag:%d---User:%d rating movie:%d with %f score----' %(flag,UserUID,MovieA,pre_rating))
 return pre_rating,flag

if __name__=='__main__':
 all_ratings=loadData()
 movie_num=100#控制電影數(shù)，只針對(duì)電影ID在該范圍的數(shù)據(jù)進(jìn)行計(jì)算，否則數(shù)據(jù)量太大 
 ratings=all_ratings[all_ratings['MovieID']<=movie_num]

 movies_id,movies_avg_rating,movies_mean=avgRating(ratings)
 pc_dic=calculatePC(ratings)#電影相似度矩陣
 #預(yù)測(cè)
 UserUID=10#當(dāng)前數(shù)據(jù)集只看過電影4，7，
 MovieA=6 
 pre_rating,flag=hadSeenMovieByUser(UserUID,MovieA,ratings,pc_dic,movies_mean)

 "-----------------測(cè)試ID提取------------------"
 #選取UserUID
 ratings.head(10)#從前10行中隨機(jī)選取一個(gè)用戶ID,例如：UserID=10
 #查看該用戶在當(dāng)前數(shù)據(jù)集中看過那些電影，方便選取新電影（防止選擇的是用戶已經(jīng)看過的電影）
 ratings[ratings['UserID']==10]#該用戶在當(dāng)前數(shù)據(jù)集中，只看過電影MovieID in(4，7)，則可選擇不是4，7的電影ID進(jìn)行預(yù)測(cè)，例如6.