def main(lineRate, alpha, topN, testN):
print("加载数据...")
train, test = tool.splitData(lineRate)
# train, test = tool.splitDataMock()
M, i2rI, r2iU, iStart = initData(train)
rankList = dict()
print("进行推荐...")
tN = 0
testNew = dict()
for user in test:
if tN == testN:
break
tN += 1
print("\r%d/%d" % (tN, testN), end='')
testNew[user] = test[user]
if user not in train:
rankList[user] = []
rankList[user] = recommend(r2iU[user], M, alpha, iStart, i2rI, topN,
train[user])
print()
eva = tool.Evaluator(train,
testNew,
rankList,
dataScale=lineRate,
topN=topN)
eva.show()
def splitDataNumpy(M, k, seed=1):
train, test = tool.splitData(1, M, k, seed)
m = 6040
n = 3952
trainM = mat(zeros((m, n))) # 最好使用默认float类型进行矩阵计算
for user, items in train.items():
for item in items:
trainM[user - 1, item - 1] = 1
return trainM, train, test
def itemCF(K, topItem, dataScale):
print("加载数据...")
train, test = tool.splitData(dataScale, 8, 0)
rankList = dict()
W = itemSimilartiy(train)
print("进行推荐...")
# 对test集的user进行推荐
for user in test.keys():
rankList[user] = recommendation(train, user, W, K, topItem)
print("进行评估...")
eva = tool.Evaluator(train, test, rankList, topN=K, dataScale=dataScale)
eva.show()
def userCF(userN, itemN, dataScale):
print("加载数据...")
train, test = tool.splitData(dataScale, 8, 0)
rankList = dict()
W = userSimilarity(train)
print("进行推荐...")
# 对test集的user进行推荐
for user in test.keys():
rankList[user] = recommend(user, train, W, userN, itemN)
print("进行评估...")
eva = tool.Evaluator(train, test, rankList, userN, dataScale)
eva.show()
def lfmRecommend(ratio, N, lineRate, topN=10):
print("加载数据...")
train, test = splitData(lineRate)
print("生成参数...")
P, Q = latentFactorModel(train, ratio, N)
rankList = dict()
print("进行推荐...")
for user in test.keys():
if user not in train:
rankList[user] = {}
continue
rankList[user] = recommend(user, P, Q, topN, train)
print("进行评估...")
eva = Evaluator(train, test, rankList, ratio=ratio, dataScale=lineRate, loopN=N)
eva.show()
def lfmRecommend(N, alpha, rate, topN=10):
print("加载数据...")
train, test = splitData(1)
print("生成参数...")
sortedUserItem = latentFactorModel(train, N, alpha, rate)
print("进行推荐...")
rankList = dict()
itemSet = set()
for items in train.values():
itemSet.update(items)
for items in test.values():
itemSet.update(items)
for user in test.keys():
if user not in train:
rankList[user] = {}
continue
rankList[user] = [(item, 1) for item in sortedUserItem[user]
if item not in train[user] and item in itemSet
][:topN]
print("进行评估...")
eva = Evaluator(train, test, rankList, loopN=N)
eva.show()
def RFsDP(self, hX, hy, privacy, num_learner, max_height, n_rows):
'''
每个用户本地RFsDP
'''
# 第p个用户lambda_p=Np/N
lamb = float(hX.shape[0] / n_rows)
# 第p个用户的训练集和验证集
hX_train, hy_train, hX_val, hy_val = splitData(hX, hy)
rfs_p = MyRandomForest(X=hX_train,
y=hy_train,
privacy_p=privacy,
num_learners=num_learner,
max_depth=max_height)
predRF_p = rfs_p.predict(hX_val)
# 第p个用户分类器准确度
acc = accuracy_score(hy_val, predRF_p, normalize=True)
# print "acc", acc
# 第p个用户w函数值
lamb_acc = math.exp(lamb) * acc
return lamb_acc, rfs_p
def AdaBoostDP(self, hX, hy, privacy, num_learner, height, n_rows):
'''
每个用户本地AdaboostDP
'''
# 第p个用户lambda_p=Np/N
lamb = float(hX.shape[0] / n_rows)
# 第p个用户的训练集和验证集
hX_train, hy_train, hX_val, hy_val = splitData(hX, hy)
ada_p = AdaboostDTBinaryClassifier(privacy_p=privacy,
num_learners=num_learner,
max_depth=height)
ada_p.fit(hX_train, hy_train)
# 对比论文13 2016年
# alpha_p.append(ada_p.learner_weight_)
# Ada_set_p.append(ada_p.learners_)
# 预测
predAda_p = ada_p.predict(hX_val)
# 第p个用户分类器准确度
acc = accuracy_score(hy_val, predAda_p, normalize=True)
lamb_acc = math.exp(lamb) * acc
return lamb_acc, ada_p