def main():
from recommend.data.datasets import load_100k
import random
# 数据集处理
df = load_100k(type='pd').alldata
nums_train = 10
df = df.groupby('user_id').filter(lambda x: len(x) >= nums_train + 10)
data_train = []
data_test = []
for gb, dtfm in df.groupby('user_id'):
dtfm = dtfm.reset_index(drop=True) # 去除原索引
index_sample = random.sample(range(len(dtfm)), nums_train)
data_train.append(dtfm.iloc[index_sample, :])
data_test.append(dtfm.drop(index_sample))
data_train = pd.concat(data_train)
data_test = pd.concat(data_test)
x_train = np.array(data_train[['user_id', 'item_id', 'timestamp']])
y_train = np.array(data_train['rating'])
x_test = np.array(data_test[['user_id', 'item_id', 'timestamp']])
y_test = np.array(data_test['rating'])
lrmf = ListRankMF(0.01, 10, 0.01, 800)
lrmf.fit(x_train, y_train, evals=(x_test, y_test), top=10)
if len(cusers) == 0: return 0
users_i = self.__item_user[i]
users_j = self.__item_user[j]
sum_up = 0.
sum_l, sum_r = 0., 0.
for u in cusers:
sum_up += ((users_i[u] - self.__item_scores[i][0]) *
(users_j[u] - self.__item_scores[j][0]))
sum_l += (users_i[u] - self.__item_scores[i][0])**2
sum_r += (users_j[u] - self.__item_scores[j][0])**2
sum_lr = np.sqrt(sum_l * sum_r)
result = sum_up / sum_lr if sum_lr > 0 else 0
return result
def __commonUsers(self, i, j):
cm_users = set(self.__item_user[i].keys()) & set(
self.__item_user[j].keys())
return cm_users
if __name__ == '__main__':
from recommend.data import datasets
df = datasets.load_100k('pd').alldata
train_x, test_x, train_y, test_y = datasets.filter_split(df, 20, 20, 0.2)
ir = ItemCR(10, 'cosine', 'origin')
ir.fit(train_x, train_y)
ir.report(test_x, 10)
for u in recommend_dict:
cm_users = set(user_item[u]) & set(recommend_dict[u])
p += len(cm_users) / top_n
r += len(cm_users) / len(user_item[u])
precision = p / len(recommend_dict)
recall = r / len(recommend_dict)
print("precision=%f\nrecall=%f" % (precision, recall))
if __name__ == '__main__':
from recommend.data.datasets import load_100k
import random
import pandas as pd
# 数据集处理
df = load_100k(type='pd').alldata
nums_train = 10
df = df.groupby('user_id').filter(lambda x: len(x) >= nums_train + 10)
data_train = []
data_test = []
for gb, dtfm in df.groupby('user_id'):
dtfm = dtfm.reset_index(drop=True) # 去除原索引
index_sample = random.sample(range(len(dtfm)), nums_train)
data_train.append(dtfm.iloc[index_sample, :])
data_test.append(dtfm.drop(index_sample))
data_train = pd.concat(data_train)
data_test = pd.concat(data_test)
x_train = np.array(data_train[['user_id', 'item_id', 'timestamp']])
y_train = np.array(data_train['rating'])
x_test = np.array(data_test[['user_id', 'item_id', 'timestamp']])
y_test = np.array(data_test['rating'])
tmp1.append(tmp3)
self.H[i] *= np.array(tmp0) / np.array(tmp1)
results = []
for u in test_uiMatrix:
for i in test_uiMatrix[u]:
r = test_uiMatrix[u][i]
results.append((r - np.dot(self.W[u], self.H[i]))**2)
rmse = np.sqrt(np.sum(results) / len(testMatrix))
print("%d iter finish, rmse: %f" % (iter, rmse))
def predict(self):
pass
if __name__ == '__main__':
from recommend.data import datasets
from sklearn.model_selection import train_test_split
mv = datasets.load_100k(type='pd')
data = mv.alldata
df_train, df_test = train_test_split(data, test_size=0.2, random_state=0)
uidset = set(df_test.user_id) - set(df_train.user_id)
if uidset:
df_test = df_test[~df_test["user_id"].isin(uidset)]
iidset = set(df_test.item_id) - set(df_train.item_id)
if iidset:
df_test = df_test[~df_test["item_id"].isin(iidset)]
nmf = NMF(10, 0.1, 20)
nmf.fit(df_train, df_test, "rmse")