def fillMatrix(F_list, step_list, LearnRating_list, penalty_list, N = 30):
trainDataSet = FO.readDataSet('trainDataSet.txt')
testDataSet = FO.readDataSet('testDataSet.txt')
userList = FO.readUserList('data/users.txt')
itemList = FO.readItemList('data/movies.txt') # 获取物品列表
dataSet = {**trainDataSet, **testDataSet}
mu = calMu(dataSet)
for F in F_list:
for step in step_list:
for LearnRating in LearnRating_list:
for penalty in penalty_list:
print(F,step,LearnRating,penalty)
path = 'SVD++/Matrix/' + str(F) + '-' + str(step) + '-' + str(LearnRating) + '-' + str(penalty)
if not os.path.exists(path):
os.makedirs(path)
bu, bi, p, q = ReadParameter(F, step, LearnRating, penalty)
for user in userList:
print(user)
rank = dict()
userRated = set([user_item[1] for user_item, value in dataSet.items() if user_item[0] == user])
UnRatedList = itemList - userRated
for item in UnRatedList:
rating = Predict(user, item, p, q, bu, bi, mu)
if math.isnan(float(rating)):
continue
rank[item] = round(rating)
if len(rank) == 0:
continue
if N > len(UnRatedList):
N = len(UnRatedList)
chooseList = random.sample(UnRatedList, N)
with open(path + '/new_ratings.txt', 'a') as fileObject:
for choose in chooseList:
fileObject.write(str(user) + '::' + str(choose) + '::' + str(rank[choose]) + '\n')
def calItemSimilarity(trainDataSet):
'''
:param trainDataSet: 训练数据集
:param type: 以何种方式计算相似度
:param simMeas: 计算相似的方法
:return: None
最后将用户之间的相似度写到文件里面
'''
# 建立 用户 到 物品 的倒排表
user_items = dict()
for user, item in trainDataSet:
if user not in user_items:
user_items[user] = set()
user_items[user].add(item)
#for user, item in user_items.items():
# print(user, ':', item)
# for item, user in item_users.items():
# print(item, ':', user)
# 读取用户列表
itemList = FO.readItemList('data/movies.txt')
#ItemRatingMean = FO.ReadRatingMean('ItemCF/ItemMean/add_trainDataSet_mean.txt')
UserRatingMean = FO.ReadRatingMean('ItemCF/UserMean/trainDataSet_mean.txt')
# 计算用户之间的相似度
for item in itemList:
W = ItemSimilarity(trainDataSet, user_items, itemList, UserRatingMean, item)
print(item)
filename = 'ItemCF/Similarity/'+str(item)+'.txt'
FO.WirteSimilarty(filename,W)
def testSystem(trainDataSet, testDataSet, K=20, N=30, type='implicit'):
testUserList = set() # 记录测试数据集里面的用户列表
for user_item, value in testDataSet.items():
testUserList.add(user_item[0])
itemList = FO.readItemList('data/movies.txt') # 获取物品列表
recall = 0 # 计算总的召回率
precision = 0 # 计算总的精确度
coverage_item = set() # 计算覆盖的物品
userLove = dict()
for user in testUserList: # 遍历所有测试用户
#userLove[user] = [(user_item[1],value) for user_item, value in trainDataSet.items() if user_item[0] == user]
userLove[user] = set([
user_item[1] for user_item, value in trainDataSet.items()
if user_item[0] == user
])
rank = Recommend(trainDataSet, itemList, userLove[user], user, K, type)
print(rank)
if N < len(rank):
top = N
else:
top = len(rank)
topRank = sorted(rank.items(),
key=operator.itemgetter(1),
reverse=True)[:top] # 获取排名前N个的推荐物品
topItemRank = [item_rating[0]
for item_rating in topRank] # 获取排名前N个的推荐物品
User_item_InTest = [
user_item[1] for user_item, rating in testDataSet.items()
if user_item[0] == user
] # 获取用户在测试集里面喜欢的物品
#print(User_item_InTest)
#print(topRank)
#print(topItemRank)
# 计算 准确率 以及 召回率
oneRecall = Recall(User_item_InTest, topItemRank)
onePrecision = Precision(User_item_InTest, topItemRank)
recall += oneRecall
precision += onePrecision
print('recall=%.4f,precision=%.4f' % (oneRecall, onePrecision))
# 计算覆盖率
Coverage(coverage_item, topItemRank)
coverage = float(len(coverage_item)) / len(itemList)
popularity = Popularity(trainDataSet, coverage_item)
recall = recall / float(len(testUserList))
precision = precision / float(len(testUserList))
print('total recall=%.4f,precision=%.4f,coverage=%.4f,popularity=%.4f'%\
(recall,precision,coverage,popularity))
return recall, precision, coverage, popularity
def calItemRatingMean(dataSet):
itemList = FO.readItemList('data/movies.txt') # 获取物品列表
itemMeanDict = dict()
for item in itemList:
print(item)
itemRating = [value for user_item, value in dataSet.items() if user_item[1] == item]
if len(itemRating) == 0:
value = 0
else:
value = sum(itemRating) / len(itemRating)
itemMeanDict[item] = value
return itemMeanDict
def calItemSimilarity(trainDataSet, type='implicit', simMeas=cosSim):
'''
:param trainDataSet: 训练数据集
:param type: 以何种方式计算相似度
:param simMeas: 计算相似的方法
:return: None
最后将用户之间的相似度写到文件里面
'''
# 建立 用户 到 物品 的倒排表
users_item = dict()
for user, item in trainDataSet:
if user not in users_item:
users_item[user] = set()
users_item[user].add(item)
for user, item in users_item.items():
print(user, ':', item)
# for item, user in item_users.items():
# print(item, ':', user)
# 读取用户列表
itemList = FO.readItemList('data/movies.txt')
# 遍历每一个用户, 计算用户之间的相似度
for item in itemList:
print(item)
if type == 'implicit': # 判读计算相似度的方式, 这里隐式计算
W = ItemSimilarity_implicit(users_item, item)
filename = 'ItemCF/ItemSimilarity/implicit/' + str(item) + '.txt'
elif type == 'explicit': # 判读计算相似度的方式, 这里显式方式
W = ItemSimilarity_explicit(trainDataSet, users_item, itemList,
item, simMeas)
filename = 'ItemCF/ItemSimilarity/explicit/' + str(
simMeas.__name__) + '/' + str(item) + '.txt'
# 将结果写入文件之中
with open(filename, 'w') as fileObject:
for users, values in W.items():
for u in users:
fileObject.write(str(str(u) + '::'))
fileObject.write(str(values) + '\n')
W = FO.readItemSimilarity(item, type) # 获取当前用户与其他用户的相似度
topItem = dict(
sorted(W.items(), key=operator.itemgetter(1),
reverse=True)[:K])
topSimItem = set([items[1] for items, value in topItem.items()])
topLoveItem = userLove & topSimItem
if len(topLoveItem) == 0:
continue
# print(topLoveItem)
for LoveItem in topLoveItem:
wji = W[item, LoveItem]
rui = dataSet[user, LoveItem]
rank[item] = rank.get(item, 0) + wji * rui
return dict(
sorted(rank.items(), key=operator.itemgetter(1), reverse=True)[:N])
if __name__ == '__main__':
trainDataSet = FO.readDataSet('ItemCF/trainDataSet.txt')
testDataSet = FO.readDataSet('ItemCF/testDataSet.txt')
'''
calItemSimilarity(trainDataSet, 'implicit') # 计算隐式相似度
calItemSimilarity(trainDataSet, 'explicit', cosSim) # 以余弦相似度计算显式相似度
calItemSimilarity(trainDataSet, 'explicit', ecludSim) # 以欧氏距离计算显式相似度
'''
itemList = FO.readItemList('data/movies.txt')
for i in range(1, 11):
rank = predictSystem(trainDataSet, testDataSet, i, itemList)
print(rank)