本例中使用得是著名得电影数据集MovieLens-100数据集
MoviesLens数据集是实现和测试电影推荐最常用得数据集之一，包含943个用户为精选得1682部电影给出得100000个电影评分
主要文件如下1:u.data 2:u.item 3:u.user
1：查看用户/电影排名信息得代码如下
import pandas as pd heads=['user_id','item_id','rating','timestamp'] ratings=pd.read_csv(r'u.data',sep='\t',names=heads) print(ratings) print("用户数量",len(ratings))
2：查看导入的电影数据表
代码如下
import pandas as pd u_cols=['user_id','age','sex','occupation','zip_code'] users=pd.read_csv(r'u.data',sep='|',names=u_cols,encoding='latin-1') print(users) r_cols=['user_id','movie_id','rating','unix_timestamp'] ratings=pd.read_csv(r'u.data',sep='\t',names=r_cols,encoding='latin-1') print(ratings) m_cols=['movie_id','title','release_data','video_release_data','imdb_url'] movies=pd.read_csv(r'u.item',sep='|',names=m_cols,usecols=range(5),encoding='latin-1') print(movies)
3：用协同过滤推荐算法进行电影推荐
误差评估如下
全部代码如下：
import pandas as pd import numpy as np from sklearn.metrics.pairwise import pairwise_distances np.set_printoptions(suppress=True) # 取消科学计数法输出 pd.set_option('display.max_rows', None) # 展示所有行 pd.set_option('display.max_columns', None) # 展示所有列 def predict(scoredata,similarity,type='user'): #基于物品得推荐 if type=='item': predt_mat=scoredata.dot(similarity)/np.array([np.abs(similarity).sum(axis=1)]) elif type=='user': #计算用户评分值 减少用户评分高低习惯影响 user_meanscorse=scoredata.mean(axis=1) score_diff=(scoredata-user_meanscorse.reshape(-1,1)) predt_mat=user_meanscorse.reshape(-1,1)+similarity.dot(score_diff)/np.array([np.abs(similarity).sum(axis=1)]).T return predt_mat #读取数据 print('step 1 读取数据') r_cols=['user_id','movie_id','rating','unix_timestamp'] scoredata=pd.read_csv(r'u.data',sep='\t',names=r_cols,encoding='latin-1') print('数据形状',scoredata.shape) #生成用户-物品评分矩阵 print('step2 生成 用户物品评分矩阵') n_users=943 n_items=1682 data_matrix=np.zeros((n_users,n_items)) for line in range(np.shape(scoredata)[0]): row=scoredata['user_id'][line]-1 col=scoredata['movie_id'][line]-1 score=scoredata['rating'][line] data_matrix[row,col]=score print('用户物品矩阵形状',data_matrix.shape) #计算相似度 print('step3 计算相似度') user_similaritry=pairwise_distances(data_matrix,metric='cosine') item_similarity=pairwise_distances(data_matrix.T,metric='cosine') print('user similarity',user_similaritry.shape) print('item similartity',item_similarity.shape) #进行相似度进行预测 print('step4 预测') user_prediction=predict(data_matrix,user_similaritry,type='user') item_perdiction=predict(data_matrix,item_similarity,type='item') #显示推荐结果 print('step 5 显示推荐结果') print('----------------') print('ubcf预测形状',user_prediction.shape) print('real answer\n',data_matrix[:5,5]) print('预测结果\n',user_prediction) print('ibcf预测形状',item_perdiction.shape) print('real answer\n',data_matrix[:5,:5]) print('预测结果\n',item_perdiction) #性能评估 print('step 6 性能评估') from sklearn.metrics import mean_squared_error from math import sqrt def rmse(predct,realNum): predct=predct[realNum.nonzero()].flatten() realNum=realNum[realNum.nonzero()].flatten() return sqrt(mean_squared_error(predct,realNum)) print('u-base mse=',str(rmse(user_prediction,data_matrix))) print('m-based mse=',str(rmse(item_perdiction,data_matrix)))
