把esmm 预测后的日记队列数量由500改成1000

eda/esmm/Model_pipline/feature_engineering.py

@@ -104,15 +104,15 @@ def feature_engineer():
 
     unique_values = []
     db = pymysql.connect(host='172.16.40.158', port=4000, user='root', passwd='3SYz54LS9#^9sBvC', db='jerry_test')
-    sql = "select distinct stat_date from esmm_train_data_share"
+    sql = "select distinct stat_date from esmm_train_data_dwell"
     unique_values.extend(get_unique(db,sql))
 
     db = pymysql.connect(host='172.16.40.158', port=4000, user='root', passwd='3SYz54LS9#^9sBvC', db='jerry_test')
-    sql = "select distinct ucity_id from esmm_train_data_share"
+    sql = "select distinct ucity_id from esmm_train_data_dwell"
     unique_values.extend(get_unique(db, sql))
 
     db = pymysql.connect(host='172.16.40.158', port=4000, user='root', passwd='3SYz54LS9#^9sBvC', db='jerry_test')
-    sql = "select distinct ccity_name from esmm_train_data_share"
+    sql = "select distinct ccity_name from esmm_train_data_dwell"
     unique_values.extend(get_unique(db, sql))
 
     db = pymysql.connect(host='172.16.40.158', port=4000, user='root', passwd='3SYz54LS9#^9sBvC', db='jerry_test')
@@ -162,16 +162,15 @@ def feature_engineer():
     # unique_values.append("video")
 
     db = pymysql.connect(host='172.16.40.158', port=4000, user='root', passwd='3SYz54LS9#^9sBvC', db='jerry_test')
-    sql = "select max(stat_date) from esmm_train_data_share"
+    sql = "select max(stat_date) from esmm_train_data_dwell"
     validate_date = con_sql(db, sql)[0].values.tolist()[0]
     print("validate_date:" + validate_date)
     temp = datetime.datetime.strptime(validate_date, "%Y-%m-%d")
     start = (temp - datetime.timedelta(days=180)).strftime("%Y-%m-%d")
     print(start)
-    print("这是分享数据")
 
     db = pymysql.connect(host='172.16.40.158', port=4000, user='root', passwd='3SYz54LS9#^9sBvC')
-    sql = "select distinct doctor.hospital_id from jerry_test.esmm_train_data_share e " \
+    sql = "select distinct doctor.hospital_id from jerry_test.esmm_train_data_dwell e " \
           "left join eagle.src_zhengxing_api_service service on e.diary_service_id = service.id " \
           "left join eagle.src_zhengxing_api_doctor doctor on service.doctor_id = doctor.id " \
           "where e.stat_date >= '{}'".format(start)
@@ -191,89 +190,92 @@ def feature_engineer():
                       29 + apps_number + level2_number + level3_number + len(unique_values)))
     value_map = dict(zip(unique_values, temp))
 
-    # sql = "select e.y,e.z,e.s,e.stat_date,e.ucity_id,feat.level2_ids,e.ccity_name,u.device_type,u.manufacturer," \
-    #       "u.channel,c.top,cut.time,dl.app_list,feat.level3_ids,doctor.hospital_id," \
-    #       "wiki.tag as tag1,question.tag as tag2,search.tag as tag3,budan.tag as tag4," \
-    #       "ot.tag as tag5,sixin.tag as tag6,cart.tag as tag7,doris.search_tag2,doris.search_tag3," \
-    #       "k.treatment_method,k.price_min,k.price_max,k.treatment_time,k.maintain_time,k.recover_time," \
-    #       "e.device_id,e.cid_id " \
-    #       "from jerry_test.esmm_train_data_share e left join jerry_test.user_feature u on e.device_id = u.device_id " \
-    #       "left join jerry_test.cid_type_top c on e.device_id = c.device_id " \
-    #       "left join jerry_test.cid_time_cut cut on e.cid_id = cut.cid " \
-    #       "left join jerry_test.device_app_list dl on e.device_id = dl.device_id " \
-    #       "left join jerry_test.diary_feat feat on e.cid_id = feat.diary_id " \
-    #       "left join jerry_test.knowledge k on feat.level2 = k.level2_id " \
-    #       "left join jerry_test.wiki_tag wiki on e.device_id = wiki.device_id " \
-    #       "left join jerry_test.question_tag question on e.device_id = question.device_id " \
-    #       "left join jerry_test.search_tag search on e.device_id = search.device_id " \
-    #       "left join jerry_test.budan_tag budan on e.device_id = budan.device_id " \
-    #       "left join jerry_test.order_tag ot on e.device_id = ot.device_id " \
-    #       "left join jerry_test.sixin_tag sixin on e.device_id = sixin.device_id " \
-    #       "left join jerry_test.cart_tag cart on e.device_id = cart.device_id " \
-    #       "left join eagle.src_zhengxing_api_service service on e.diary_service_id = service.id " \
-    #       "left join eagle.src_zhengxing_api_doctor doctor on service.doctor_id = doctor.id " \
-    #       "left join jerry_test.search_doris doris on e.device_id = doris.device_id and e.stat_date = doris.get_date " \
-    #       "where e.stat_date >= '{}'".format(start)
-    #
-    # df = spark.sql(sql)
-    #
-    # df = df.drop_duplicates(["ucity_id", "level2_ids", "ccity_name", "device_type", "manufacturer",
-    #                          "channel", "top", "time", "stat_date", "app_list", "hospital_id", "level3_ids",
-    #                          "tag1", "tag2", "tag3", "tag4", "tag5", "tag6", "tag7"])
-    #
-    # df = df.na.fill(dict(zip(features, features)))
-    #
-    # rdd = df.select("stat_date", "y", "z", "app_list", "level2_ids", "level3_ids",
-    #                 "tag1", "tag2", "tag3", "tag4", "tag5", "tag6", "tag7",
-    #                 "ucity_id", "ccity_name", "device_type", "manufacturer", "channel", "top", "time",
-    #                 "hospital_id", "treatment_method", "price_min", "price_max", "treatment_time",
-    #                 "maintain_time", "recover_time", "search_tag2", "search_tag3","cid_id","device_id","s")\
-    #     .rdd.repartition(200).map(
-    #     lambda x: (x[0], float(x[1]), float(x[2]), app_list_func(x[3], app_list_map), app_list_func(x[4], leve2_map),
-    #                app_list_func(x[5], leve3_map), app_list_func(x[6], leve2_map), app_list_func(x[7], leve2_map),
-    #                app_list_func(x[8], leve2_map), app_list_func(x[9], leve2_map), app_list_func(x[10], leve2_map),
-    #                app_list_func(x[11], leve2_map), app_list_func(x[12], leve2_map),
-    #                [value_map.get(x[0], 1), value_map.get(x[13], 2), value_map.get(x[14], 3), value_map.get(x[15], 4),
-    #                 value_map.get(x[16], 5), value_map.get(x[17], 6), value_map.get(x[18], 7), value_map.get(x[19], 8),
-    #                 value_map.get(x[20], 9), value_map.get(x[21], 10),
-    #                 value_map.get(x[22], 11), value_map.get(x[23], 12), value_map.get(x[24], 13),
-    #                 value_map.get(x[25], 14), value_map.get(x[26], 15)],
-    #                app_list_func(x[27], leve2_map), app_list_func(x[28], leve3_map),x[13],x[29],x[30],float(x[31])
-    #                ))
-    #
-    #
-    # rdd.persist(storageLevel= StorageLevel.MEMORY_ONLY_SER)
-    #
-    # # TODO 上线后把下面train fliter 删除，因为最近一天的数据也要作为训练集
-    #
-    # train = rdd.filter(lambda x: x[0] != validate_date).map(
-    #     lambda x: (x[1], x[2], x[3], x[4], x[5], x[6], x[7], x[8], x[9],
-    #                x[10], x[11], x[12], x[13], x[14], x[15],x[16],x[17],x[18],x[19]))
-    # f = time.time()
-    # spark.createDataFrame(train).toDF("y", "z", "app_list", "level2_list", "level3_list",
-    #                                   "tag1_list", "tag2_list", "tag3_list", "tag4_list",
-    #                                   "tag5_list", "tag6_list", "tag7_list", "ids",
-    #                                   "search_tag2_list","search_tag3_list","city","cid_id","uid","s") \
-    #     .repartition(1).write.format("tfrecords").save(path=path + "tr/", mode="overwrite")
-    # h = time.time()
-    # print("train tfrecord done")
-    # print((h - f) / 60)
-    #
-    # get_pre_number()
-    #
-    # test = rdd.filter(lambda x: x[0] == validate_date).map(
-    #     lambda x: (x[1], x[2], x[3], x[4], x[5], x[6], x[7], x[8], x[9],
-    #                x[10], x[11], x[12], x[13], x[14], x[15],x[16],x[17],x[18],x[19]))
-    #
-    # spark.createDataFrame(test).toDF("y", "z", "app_list", "level2_list", "level3_list",
-    #                                  "tag1_list", "tag2_list", "tag3_list", "tag4_list",
-    #                                  "tag5_list", "tag6_list", "tag7_list", "ids",
-    #                                  "search_tag2_list","search_tag3_list","city","cid_id","uid","s") \
-    #     .repartition(1).write.format("tfrecords").save(path=path + "va/", mode="overwrite")
-    #
-    # print("va tfrecord done")
-    #
-    # rdd.unpersist()
+    sql = "select e.y,e.z,e.stat_date,e.ucity_id,feat.level2_ids,e.ccity_name,u.device_type,u.manufacturer," \
+          "u.channel,c.top,cut.time,dl.app_list,feat.level3_ids,doctor.hospital_id," \
+          "wiki.tag as tag1,question.tag as tag2,search.tag as tag3,budan.tag as tag4," \
+          "ot.tag as tag5,sixin.tag as tag6,cart.tag as tag7,doris.search_tag2,doris.search_tag3," \
+          "k.treatment_method,k.price_min,k.price_max,k.treatment_time,k.maintain_time,k.recover_time," \
+          "e.device_id,e.cid_id " \
+          "from jerry_test.esmm_train_data_dwell e left join jerry_test.user_feature u on e.device_id = u.device_id " \
+          "left join jerry_test.cid_type_top c on e.device_id = c.device_id " \
+          "left join jerry_test.cid_time_cut cut on e.cid_id = cut.cid " \
+          "left join jerry_test.device_app_list dl on e.device_id = dl.device_id " \
+          "left join jerry_test.diary_feat feat on e.cid_id = feat.diary_id " \
+          "left join jerry_test.knowledge k on feat.level2 = k.level2_id " \
+          "left join jerry_test.wiki_tag wiki on e.device_id = wiki.device_id " \
+          "left join jerry_test.question_tag question on e.device_id = question.device_id " \
+          "left join jerry_test.search_tag search on e.device_id = search.device_id " \
+          "left join jerry_test.budan_tag budan on e.device_id = budan.device_id " \
+          "left join jerry_test.order_tag ot on e.device_id = ot.device_id " \
+          "left join jerry_test.sixin_tag sixin on e.device_id = sixin.device_id " \
+          "left join jerry_test.cart_tag cart on e.device_id = cart.device_id " \
+          "left join eagle.src_zhengxing_api_service service on e.diary_service_id = service.id " \
+          "left join eagle.src_zhengxing_api_doctor doctor on service.doctor_id = doctor.id " \
+          "left join jerry_test.search_doris doris on e.device_id = doris.device_id and e.stat_date = doris.get_date " \
+          "where e.stat_date >= '{}'".format(start)
+
+    df = spark.sql(sql)
+
+    df = df.drop_duplicates(["ucity_id", "level2_ids", "ccity_name", "device_type", "manufacturer",
+                             "channel", "top", "time", "stat_date", "app_list", "hospital_id", "level3_ids",
+                             "tag1", "tag2", "tag3", "tag4", "tag5", "tag6", "tag7"])
+
+    df = df.na.fill(dict(zip(features, features)))
+
+    rdd = df.select("stat_date", "y", "z", "app_list", "level2_ids", "level3_ids",
+                    "tag1", "tag2", "tag3", "tag4", "tag5", "tag6", "tag7",
+                    "ucity_id", "ccity_name", "device_type", "manufacturer", "channel", "top", "time",
+                    "hospital_id", "treatment_method", "price_min", "price_max", "treatment_time",
+                    "maintain_time", "recover_time", "search_tag2", "search_tag3","cid_id","device_id")\
+        .rdd.repartition(200).map(
+        lambda x: (x[0], float(x[1]), float(x[2]), app_list_func(x[3], app_list_map), app_list_func(x[4], leve2_map),
+                   app_list_func(x[5], leve3_map), app_list_func(x[6], leve2_map), app_list_func(x[7], leve2_map),
+                   app_list_func(x[8], leve2_map), app_list_func(x[9], leve2_map), app_list_func(x[10], leve2_map),
+                   app_list_func(x[11], leve2_map), app_list_func(x[12], leve2_map),
+                   [value_map.get(x[0], 1), value_map.get(x[13], 2), value_map.get(x[14], 3), value_map.get(x[15], 4),
+                    value_map.get(x[16], 5), value_map.get(x[17], 6), value_map.get(x[18], 7), value_map.get(x[19], 8),
+                    value_map.get(x[20], 9), value_map.get(x[21], 10),
+                    value_map.get(x[22], 11), value_map.get(x[23], 12), value_map.get(x[24], 13),
+                    value_map.get(x[25], 14), value_map.get(x[26], 15)],
+                   app_list_func(x[27], leve2_map), app_list_func(x[28], leve3_map),x[13],x[29],x[30]
+                   ))
+
+
+    rdd.persist(storageLevel= StorageLevel.MEMORY_ONLY_SER)
+
+    # TODO 上线后把下面train fliter 删除，因为最近一天的数据也要作为训练集
+
+    train = rdd.map(
+        lambda x: (x[1], x[2], x[3], x[4], x[5], x[6], x[7], x[8], x[9],
+                   x[10], x[11], x[12], x[13], x[14], x[15],x[16],x[17],x[18]))
+    f = time.time()
+    spark.createDataFrame(train).toDF("y", "z", "app_list", "level2_list", "level3_list",
+                                      "tag1_list", "tag2_list", "tag3_list", "tag4_list",
+                                      "tag5_list", "tag6_list", "tag7_list", "ids",
+                                      "search_tag2_list","search_tag3_list","city","cid_id","uid") \
+        .repartition(1).write.format("tfrecords").save(path=path + "tr/", mode="overwrite")
+    h = time.time()
+    print("train tfrecord done")
+    print((h - f) / 60)
+
+    print("训练集样本总量：")
+    print(rdd.count())
+
+    get_pre_number()
+
+    test = rdd.filter(lambda x: x[0] == validate_date).map(
+        lambda x: (x[1], x[2], x[3], x[4], x[5], x[6], x[7], x[8], x[9],
+                   x[10], x[11], x[12], x[13], x[14], x[15],x[16],x[17],x[18]))
+
+    spark.createDataFrame(test).toDF("y", "z", "app_list", "level2_list", "level3_list",
+                                     "tag1_list", "tag2_list", "tag3_list", "tag4_list",
+                                     "tag5_list", "tag6_list", "tag7_list", "ids",
+                                     "search_tag2_list","search_tag3_list","city","cid_id","uid") \
+        .repartition(1).write.format("tfrecords").save(path=path + "va/", mode="overwrite")
+
+    print("va tfrecord done")
+
+    rdd.unpersist()
 
     return validate_date, value_map, app_list_map, leve2_map, leve3_map
 
@@ -285,7 +287,7 @@ def get_predict(date,value_map,app_list_map,leve2_map,leve3_map):
           "wiki.tag as tag1,question.tag as tag2,search.tag as tag3,budan.tag as tag4," \
           "ot.tag as tag5,sixin.tag as tag6,cart.tag as tag7,doris.search_tag2,doris.search_tag3," \
           "k.treatment_method,k.price_min,k.price_max,k.treatment_time,k.maintain_time,k.recover_time " \
-          "from jerry_test.esmm_data e " \
+          "from jerry_test.esmm_pre_data e " \
           "left join jerry_test.user_feature u on e.device_id = u.device_id " \
           "left join jerry_test.cid_type_top c on e.device_id = c.device_id " \
           "left join jerry_test.cid_time_cut cut on e.cid_id = cut.cid " \
@@ -299,8 +301,7 @@ def get_predict(date,value_map,app_list_map,leve2_map,leve3_map):
           "left join jerry_test.sixin_tag sixin on e.device_id = sixin.device_id " \
           "left join jerry_test.cart_tag cart on e.device_id = cart.device_id " \
           "left join jerry_test.knowledge k on feat.level2 = k.level2_id " \
-          "left join jerry_test.search_doris doris on e.device_id = doris.device_id and e.stat_date = doris.get_date " \
-          "limit 20000"
+          "left join jerry_test.search_doris doris on e.device_id = doris.device_id and e.stat_date = doris.get_date"
 
     features = ["ucity_id", "ccity_name", "device_type", "manufacturer",
                 "channel", "top", "time", "hospital_id",
@@ -388,9 +389,3 @@ if __name__ == '__main__':
     get_predict(validate_date, value_map, app_list_map, leve2_map, leve3_map)
 
     spark.stop()
-
-
-
-
-
-

eda/esmm/Model_pipline/train.py

@@ -378,7 +378,7 @@ def trans(x):
 def set_join(lst):
     l = lst.unique().tolist()
     r = [str(i) for i in l]
-    r =r[:500]
+    r =r[:1000]
     return ','.join(r)
 
 def df_sort(result,queue_name):