机器学习原理与实战_源代码和实验数据.rar

# 代码10-9 import pandas as pd import numpy as np media3 = pd.read_csv('../tmp/media3.csv', header='infer', error_bad_lines=False) # 构建家庭成员标签 live_label = pd.read_csv('../data/table_livelabel.csv', encoding='gbk') # 时间列存在很多种写法，而且存在隔天的情况 live_label.开始时间 = pd.to_datetime(live_label.开始时间) # 将时间列变成datetime类型，好比较大小 live_label.结束时间 = pd.to_datetime(live_label.结束时间) live_label['origin_time1'] = live_label.开始时间.apply(lambda x: x.second + x.minute * 60 + x.hour * 3600) live_label['end_time1'] = live_label.结束时间.apply(lambda x: x.second + x.minute * 60 + x.hour * 3600) print('查看星期:', live_label.星期.unique()) # 有些节目跨夜，需进行隔夜处理 def geyechuli_xingqi(x): dic = {'星期一':'星期二', '星期二':'星期三', '星期三':'星期四', '星期四':'星期五', '星期五':'星期六', '星期六':'星期日', '星期日':'星期一'} return x.apply(lambda y: dic[y.星期], axis=1) ind1 = live_label.结束时间 < live_label.开始时间 label1 = live_label.loc[ind1, :].copy() # 日期可以变，后面以end_time比较 live_label.loc[ind1, '结束时间'] = pd.Timestamp('2018-06-07 23:59:59') live_label.loc[ind1, 'end_time1'] = 24 * 3600 label1.iloc[:, 1] = pd.Timestamp('2018-06-07 00:00:00') label1.iloc[:, -2] = 0 label1.iloc[:, 0] = geyechuli_xingqi(label1) label = pd.concat([live_label, label1]) label = label.reset_index(drop = True) # 恢复默认索引 data_pindao = media3.copy() label_ = label.loc[:, ['星期', 'origin_time1', 'end_time1', '频道', '适用人群']] label_.columns = ['星期', 'origin_time1', 'end_time1', 'station_name', '适用人群'] media_ = data_pindao.loc[:, ['phone_no', '星期', 'origin_time1', 'end_time1', 'station_name', ]] family_ = pd.merge(media_, label_, how = 'left', on=['星期', 'station_name']) f = np.array(family_.loc[:, ['origin_time1_x', 'end_time1_x', 'origin_time1_y', 'end_time1_y']]) # lebel中的栏目记录分为四类：一类是只看了后半截，一类是全部都看了， # 一类是只看了前半截，一类是看了中间一截 n1 = np.apply_along_axis(lambda x: (x[0] > x[2])&(x[0] < x[3])&(x[1] >= x[3]) , 1, f) # 1是行，2是列 n2 = np.apply_along_axis(lambda x: ((x[0] <= x[2])&(x[1] >= x[3])) , 1, f) n3 = np.apply_along_axis(lambda x: ((x[1] > x[2])&(x[1] < x[3])&(x[0] <=x [2])), 1, f) n4 = np.apply_along_axis(lambda x: ((x[0] > x[2])&(x[1] < x[3])), 1, f) da1 = family_.loc[n1, :].copy() da1['wat_time'] = da1.end_time1_y - da1.origin_time1_x da2 = family_.loc[n2, :].copy() da2['wat_time'] = da2.end_time1_y - da2.origin_time1_y da3 = family_.loc[n3, :].copy() da3['wat_time'] = da3.end_time1_x - da3.origin_time1_y da4= family_.loc[n4, :].copy() da4['wat_time'] = da4.end_time1_x - da4.origin_time1_x sd = pd.concat([da1, da2, da3, da4]) grouped = pd.DataFrame(sd['wat_time'].groupby([sd['phone_no'], sd['适用人群']]).sum()) grouped1 = pd.DataFrame(data_pindao['wat_time'].groupby([data_pindao['phone_no']]).sum()) phone_no = [] for i in range(len(grouped)): id = grouped.index[i][0] if id in grouped1.index.unique(): shang = grouped['wat_time'][i] / grouped1[grouped1.index==id] if shang.values > 0.16: phone_no.append(grouped.index[i][0]) else: continue grouped2 = grouped.reset_index() # 找出满足0.16标准的用户的家庭成员 aaa = pd.DataFrame(np.zeros([0, 3]), columns = grouped2.columns) for k in phone_no: aaa = pd.concat([aaa, grouped2.ix[grouped2.iloc[:, 0]== k, :]], axis=0) a = [aaa.ix[aaa['phone_no'] == k, '适用人群'].tolist() for k in aaa['phone_no'].unique()] a = pd.Series([pd.Series(a[i]).unique() for i in range(len(a))]) a = pd.DataFrame(a) b = pd.DataFrame(aaa['phone_no'].unique()) c = pd.concat([a, b], axis=1) c.columns = ['家庭成员', 'phone_no'] grouped1 = grouped1.reset_index() users_label = pd.merge(grouped1, c, left_on='phone_no', right_on ='phone_no', how='left') # 构建电视依赖度标签 di = media3.phone_no.value_counts().values < 10 users_label['电视依赖度'] = 0 users_label.loc[di, '电视依赖度'] = '低' zhong_gao = [i for i in users_label.index if i not in di] num = media3.phone_no.value_counts() for i in zhong_gao: if (users_label.loc[i, 'wat_time'] / num.iloc[i]) <= 3000: users_label.loc[i, '电视依赖度'] = '中' users_label.loc[users_label.电视依赖度 == 0, '电视依赖度'] = '高' # 构建机顶盒名称标签 jidinghe = media3.ix[media3['res_type'] == 1, :] jdh = jidinghe.res_name.groupby(jidinghe.phone_no).unique() jdh = jdh.reset_index() jdh.columns = ['phone_no', '机顶盒名称'] users_label = pd.merge(users_label, jdh, left_on='phone_no', right_on ='phone_no', how='left') # 观看时间偏好（周末） media_watch = media3.loc[:, ['phone_no', 'origin_time', 'end_time', 'res_type', '星期', 'wat_time']] media_f1 = media_watch.ix[media_watch['星期'] == '星期六', :] media_f2 = media_watch.ix[media_watch['星期'] == '星期日', :] media_freeday = pd.concat([media_f1, media_f2], axis=0) media_freeday = media_freeday.reset_index(drop = True) # 恢复默认索引 ''' 由于观看时间段偏好（工作日）与观看时间偏好（周末）的计算方式相似， 所以此处不在列出观看时间段偏好（工作日）的计算代码 ''' # 分割日期和时间，按空格号分开 T1 = [str(media_freeday.ix[i, 1]).split(' ') for i in list(media_freeday.index)] # T1是列表，time[i] = T1[[i]][2]表示T1中第i个列表的第二列赋值给time的第i个 media_freeday['origin_time'] = [' '.join(['2018/06/09', T1[i][1]]) for i in media_freeday.index] media_freeday['origin_time'] = pd.to_datetime(media_freeday['origin_time'], format = '%Y/%m/%d %H:%M') point = ['2018/06/09 00:00:00', '2018/06/09 06:00:00', '2018/06/09 09:00:00', '2018/06/09 11:00:00', '2018/06/09 14:00:00', '2018/06/09 16:00:00', '2018/06/09 18:00:00', '2018/06/09 22:00:00', '2018/06/09 23:59:59'] lab = ['凌晨', '早晨', '上午', '中午', '下午', '傍晚', '晚上', '深夜'] sjd_num = pd.DataFrame() for k in range(0, 8): kk = (media_freeday['origin_time'] >= point[k]) & \ (media_freeday['origin_time'] < point[k+1]) sjd = media_freeday.ix[kk==True, ['phone_no', 'wat_time']] sjd_new = sjd.groupby('phone_no').sum().sort_values('wat_time') sjd_new['时间段偏好（周末）'] = lab[k] sjd_num = pd.concat([sjd_num, sjd_new], axis=0) sjd_num = sjd_num.reset_index() # 增加索引 sjd_num = sjd_num.sort_values('phone_no') # 以用户号排序 sjd_num = sjd_num.reset_index(drop = True) # 增加默认索引 # 保留前3的标签 users = sjd_num['phone_no'].unique() sjd_num_new = pd.DataFrame() for m in users: gd = sjd_num.ix[sjd_num['phone_no'] == m, :] if len(gd)>3: gd = gd.sort_values('wat_time').iloc[::-1, :] gd = gd.iloc[:3, :] else: continue sjd_num_new = pd.concat([sjd_num_new, gd], axis=0) sjd_label = sjd_num_new['时间段偏好（周末）'].groupby(sjd_num_new['phone_no']).sum() sjd_label = sjd_label.reset_index() # 增加索引 users_label = pd.merge(users_label, sjd_label, left_on='phone_no', right_on ='phone_no', how='left') # 构建付费频道月均收视时长标签 import re ffpd_ind =[re.search('付费', str(i))!=None for i in media3.ix[:, 'station_name']] media_ffpd = media3.ix[ffpd_ind, :] ffpd = media_ffpd['wat_time'].groupby(media_ffpd['phone_no']).sum() ffpd = ffpd.reset_index() # 增加索引 ffpd['付费频道月均收视时长'] = 0 for i in range(len(ffpd)): if ffpd.iloc[i, 1]