BayesCombine.py 13 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267
  1. import yaml
  2. import os
  3. import pandas as pd
  4. import numpy as np
  5. from utils.commonFunc import update_dict, get_db_engine, city_code_transform, age_code_transform, gender_code_transform
  6. from functools import reduce
  7. from itertools import product
  8. from datetime import datetime
  9. from statsmodels.stats.proportion import proportions_ztest
  10. import traceback
  11. import logging
  12. from concurrent_log import ConcurrentTimedRotatingFileHandler
  13. log_formatter = logging.Formatter('%(asctime)s %(levelname)s %(message)s', '%Y/%m/%d %I:%M:%S %p')
  14. log_handler = ConcurrentTimedRotatingFileHandler("logs/BayesCombine.log", when="midnight", backupCount=100)
  15. log_handler.setFormatter(log_formatter)
  16. logger = logging.getLogger('bayes_combine_logger')
  17. logger.addHandler(log_handler)
  18. logger.setLevel(logging.DEBUG)
  19. print('id of bayes_combine_logger %s' % id(logger))
  20. class BayesFeatures(object):
  21. """
  22. 计算素材单维度的特征值
  23. """
  24. def __init__(self, signature, target_type, dim, dim_config):
  25. self.signature = signature
  26. self.target_type = target_type
  27. self.dim = dim
  28. self.dim_config = dim_config
  29. self.table = self.dim_config['table']
  30. self.file_name = self.dim_config['fieldName']
  31. self.dim_lst = self.dim_config['Lst']
  32. self.window_size = self.dim_config['windowSize']
  33. self.window_combine = []
  34. self.bayes_feature = {}
  35. def get_window_combine(self):
  36. """
  37. 根据size得到指定维度的滑窗组合。
  38. 如 age: ['18-23岁', '24-30岁','31-40岁',....] 和 size = 2
  39. 得到 ['18-23岁', '24-30岁'],['24-30岁','31-40岁'],......
  40. """
  41. start = 0
  42. while (start < len(self.dim_lst)) and (start + self.window_size <= len(self.dim_lst)):
  43. self.window_combine.append(self.dim_lst[start: start + self.window_size])
  44. start += 1
  45. # 添加一个全部的组合,等同于'不限'
  46. self.window_combine.append(['不限'])
  47. def get_bayes_feature(self):
  48. # 1、获取指定素材在指定维度下的人群数据(近一个月内的数据表现)
  49. sql = '''
  50. select signature,
  51. %s,
  52. sum(aclick) aclick,
  53. sum(bclick) bclick,
  54. sum(activation) activation
  55. from %s
  56. where signature = '%s'
  57. -- and datediff(now(),stat_date)<=30
  58. group by signature, %s
  59. ''' % (self.file_name, self.table, self.signature, self.file_name)
  60. df = pd.read_sql(sql, engine)
  61. if self.target_type == 'action_ratio':
  62. # 计算行为率(bclick/aclick)的组合特征值
  63. df['unbclick'] = df['aclick'] - df['bclick']
  64. for sub_combine in self.window_combine:
  65. if sub_combine != ['不限']:
  66. key = self.dim + '_' + '|'.join(sub_combine)
  67. if self.dim != 'city':
  68. pos_pct = df[df[self.file_name].isin(sub_combine)].bclick.sum() / df.bclick.sum()
  69. neg_pct = df[df[self.file_name].isin(sub_combine)].unbclick.sum() / df.unbclick.sum()
  70. if self.dim == 'city':
  71. # 获取 city_level 对应的 city_name
  72. city_lst = []
  73. for city_level in sub_combine:
  74. get_city_sql = """
  75. select city_name from ctop_kuaishou_city_level where city_level = '%s'
  76. """ % city_level
  77. city_df = pd.read_sql(get_city_sql, engine)
  78. city_lst.extend(city_df['city_name'].values)
  79. pos_pct = df[df[self.file_name].isin(city_lst)].bclick.sum() / df.bclick.sum()
  80. neg_pct = df[df[self.file_name].isin(city_lst)].unbclick.sum() / df.unbclick.sum()
  81. self.bayes_feature[key] = {'pos_pct': pos_pct, 'neg_pct': neg_pct}
  82. else:
  83. self.bayes_feature[self.dim + '_' + '不限'] = {'pos_pct': 1, 'neg_pct': 1}
  84. if self.target_type == 'convert_ratio':
  85. # 计算转化率(activation/bclick)的组合特征值
  86. df['unactivation'] = df['bclick'] - df['activation']
  87. for sub_combine in self.window_combine:
  88. if sub_combine != '不限':
  89. key = self.dim + '_' + '|'.join(sub_combine)
  90. if self.dim != 'city':
  91. pos_pct = df[df[self.file_name].isin(sub_combine)].activation.sum() / df.activation.sum()
  92. neg_pct = df[df[self.file_name].isin(sub_combine)].unactivation.sum() / df.unactivation.sum()
  93. if self.dim == 'city':
  94. # 获取 city_level 对应的 city_name
  95. city_lst = []
  96. for city_level in sub_combine:
  97. get_city_sql = """
  98. select city_name from ctop_kuaishou_city_level where city_level = '%s'
  99. """ % city_level
  100. city_df = pd.read_sql(get_city_sql, engine)
  101. city_lst.extend(city_df['city_name'].values)
  102. pos_pct = df[df[self.file_name].isin(city_lst)].activation.sum() / df.activation.sum()
  103. neg_pct = df[df[self.file_name].isin(city_lst)].unactivation.sum() / df.unactivation.sum()
  104. self.bayes_feature[key] = {'pos_pct': pos_pct, 'neg_pct': neg_pct}
  105. else:
  106. self.bayes_feature[self.dim + '_' + '不限'] = {'pos_pct': 1, 'neg_pct': 1}
  107. class BayesCombine(object):
  108. """
  109. 依据特征值,计算多维度的组合预估值
  110. """
  111. def __init__(self, signature, project_id, target_type, dim_features):
  112. self.signature = signature
  113. self.project_id = project_id
  114. self.target_type = target_type
  115. self.dim_features = dim_features
  116. self.bayes_combine_df = pd.DataFrame()
  117. self.actual_prob = None
  118. self.sample_size = None
  119. self.sig_pos_pct = None
  120. self.sig_neg_pct = None
  121. def get_bayes_estimate(self):
  122. dim_combine_lst = [[key for key in fea.keys()] for fea in self.dim_features]
  123. self.dim_features = reduce(update_dict, self.dim_features)
  124. # 通过 年龄对应的表: ctop_kuaishou_audience_daily_report_by_signature_age 来获取样本量,和正负比例值(其他表的数据可能会丢失导致不全)
  125. get_sample_and_pct_sql = """
  126. select sum(aclick) aclick, sum(bclick) bclick, sum(activation) activation from %s where
  127. signature = '%s'
  128. -- and datediff(now(),stat_date)<=30
  129. """ % (config['bayesDim']['age']['table'],self.signature)
  130. sample_pct_df = pd.read_sql(get_sample_and_pct_sql, engine)
  131. if self.target_type == 'action_ratio':
  132. self.sig_pos_pct = sample_pct_df['bclick'].sum() / sample_pct_df['aclick'].sum()
  133. self.sig_neg_pct = 1 - self.sig_pos_pct
  134. self.actual_prob = self.sig_pos_pct
  135. self.sample_size = int(sample_pct_df.aclick.sum())
  136. elif self.target_type == 'convert_ratio':
  137. self.sig_pos_pct = sample_pct_df['activation'].sum() / sample_pct_df['bclick'].sum()
  138. self.sig_neg_pct = 1 - self.sig_pos_pct
  139. self.actual_prob = self.sig_pos_pct
  140. self.sample_size = int(sample_pct_df.bclick.sum())
  141. for ele in product(*dim_combine_lst):
  142. pos_pct_lst = [self.dim_features[key]['pos_pct'] for key in ele]
  143. prob_pos = reduce(lambda x, y: x * y, pos_pct_lst)
  144. neg_pct_lst = [self.dim_features[key]['neg_pct'] for key in ele]
  145. prob_neg = reduce(lambda x, y: x * y, neg_pct_lst)
  146. prob = (prob_pos * self.sig_pos_pct) / (prob_neg * self.sig_neg_pct)
  147. out_dict = dict(zip([e.split('_')[0] for e in ele], [e.split('_')[-1] for e in ele]))
  148. # 计算该组合的概率值 与 实际投放的概率值 是否存在显著性差异:显著好、 显著差
  149. count = np.array([self.sample_size * prob, self.sample_size * self.actual_prob])
  150. nobs = np.array([self.sample_size, self.sample_size])
  151. z_score, p_value = proportions_ztest(count=count, nobs=nobs, value=None, alternative='two-sided', prop_var=False)
  152. out_dict['z_score'] = z_score
  153. out_dict['p_value'] = p_value
  154. out_dict['sample_size'] = self.sample_size
  155. # TODO 调用人群预估覆盖接口,得到该组合的人群覆盖数
  156. # 获取该项目下在投的账号
  157. get_acc_sql = """select account_id from ctop_user_allocation where project_id = %s and account_status=0 limit 1""" % \
  158. self.project_id
  159. account_df = pd.read_sql(get_acc_sql, engine)
  160. account_id = account_df['account_id'].values[0]
  161. request_data = {'region': city_code_transform(out_dict.get('city')),
  162. 'ages_range': age_code_transform(out_dict.get('age')),
  163. 'gender': gender_code_transform(out_dict.get('gender')),
  164. 'advertiser_id': account_id}
  165. out_dict['crowd_coverage_cnt'] = 1000
  166. out_dict['combine_estimate_prob'] = prob
  167. out_dict['actual_prob'] = self.actual_prob
  168. out_dict['signature'] = self.signature
  169. out_dict['project_id'] = self.project_id
  170. out_dict['target_type'] = self.target_type
  171. out_dict['stat_date'] = str(datetime.now().date())
  172. combine = pd.DataFrame([out_dict])
  173. self.bayes_combine_df = self.bayes_combine_df.append(combine, ignore_index=True)
  174. def write_to_db(self):
  175. self.bayes_combine_df.to_sql(name="ctop_ai_kuaishou_signature_recommended_target_combine",
  176. con=engine,
  177. if_exists='append',
  178. index=False)
  179. if __name__ == '__main__':
  180. try:
  181. # 1、读取配置文件
  182. with open('config/config.yaml', mode='r', encoding='utf-8') as f:
  183. config = yaml.load(f.read(), Loader=yaml.FullLoader)
  184. # 1-1 数据库连接引擎,依据开发环境/生产环境 进行切换
  185. if os.getenv('LYY_DEV', 'unknown') == 'dev':
  186. engine = get_db_engine(config['devDB'])
  187. else:
  188. engine = get_db_engine(config['productDB'])
  189. # 2、 参与定向组合的维度
  190. target_dim = [key for key in config['bayesDim'].keys() if config['bayesDim'][key]['isOn']]
  191. # 4、计算贝叶斯组合入库
  192. for project_id in config['projectId']:
  193. # 4.1 获取指定项目下当前活跃的素材信息, 如近3天内累计激活个数达到50个(开发环境或者生产环境,这部分都读取生产数据库)
  194. product_engine = get_db_engine(config['productDB'])
  195. sql = '''
  196. select signature from ctop_kuaishou_report_daily_material
  197. where account_id in (select account_id from ctop_user_allocation where project_id = %s)
  198. and datediff(now(),stat_date) <= %s
  199. group by signature
  200. having sum(activation) >= %s
  201. ''' % (project_id,
  202. config['activeMaterialFilterRule']['days'],
  203. config['activeMaterialFilterRule']['activation'])
  204. df = pd.read_sql(sql, product_engine)
  205. active_signatures = df[~df.signature.isnull()].signature.values
  206. # 4.2 在素材人群报表筛选里面近一个月累计100个激活
  207. # TODO 等生产的素材人群表存在之后,修改表名,添加 days
  208. sql = """
  209. select signature from ctop_kuaishou_audience_daily_report_by_signature_age
  210. where signature in %s
  211. group by signature
  212. having sum(activation) >= %s
  213. """ % (tuple(active_signatures),
  214. config['getBaysCombineMaterialFilterRule']['activation'])
  215. df = pd.read_sql(sql, engine)
  216. signature_lst = df['signature'].values
  217. # TODO 这行代码用于测试
  218. signature_lst = ['0070efb7557b2a04cf3d4a6f243c3cd8', '03b93728f0d82ea7865c3c7cf632bc1b']
  219. # 4.2 计算指定素材的贝叶斯特征
  220. for sig in signature_lst:
  221. for t_type in config['targetType']:
  222. bayes_feature_lst = []
  223. for dimension in target_dim:
  224. cls = BayesFeatures(sig, t_type, dimension, config['bayesDim'][dimension])
  225. # 计算滑窗组合
  226. cls.get_window_combine()
  227. cls.get_bayes_feature()
  228. bayes_feature_lst.append(cls.bayes_feature)
  229. # 依据特征值,计算多维度的组合预估值
  230. cls = BayesCombine(sig, project_id, t_type, bayes_feature_lst)
  231. cls.get_bayes_estimate()
  232. cls.write_to_db()
  233. logger.info('project_id is %s 完成贝叶斯组合计算!' % project_id)
  234. except:
  235. logger.error("traceback is %s" % (traceback.format_exc()))