import multiprocessing
import os
import sys
import time
from collections import defaultdict
sys.path.append(os.path.realpath("."))
from gensim.models import Word2Vec, word2vec
from utils.date import get_ndays_before_no_minus, get_ndays_before_with_format
from utils.files import DATA_PATH, MODEL_PATH
from utils.spark import get_spark
tractate_click_ids_model_path = os.path.join(MODEL_PATH, "tractate_click_ids_item2vec_model")
try:
TRACTATE_CLICK_IDS_MODEL = word2vec.Word2Vec.load(tractate_click_ids_model_path)
TRACTATE_CLICK_IDS = set(TRACTATE_CLICK_IDS_MODEL.wv.vocab.keys())
except Exception as e:
print(e)
def get_tracate_click_data(spark, start, end):
reg = r"""^\\d+$"""
sql = """
SELECT DISTINCT t1.partition_date, t1.cl_id, cast(t1.business_id as int) card_id, t1.app_session_id
FROM
(select partition_date,cl_id,business_id,action,page_name,page_stay, app_session_id
from online.bl_hdfs_maidian_updates
where action = 'page_view'
AND partition_date BETWEEN '{}' AND '{}'
AND page_name='user_post_detail'
AND page_stay>=4
AND cl_id is not null
AND cl_id != ''
AND business_id is not null
AND business_id != ''
AND business_id rlike '{}'
) AS t1
JOIN
(select partition_date,active_type,first_channel_source_type,device_id
from online.ml_device_day_active_status
where partition_date BETWEEN '{}' AND '{}'
AND active_type IN ('1', '2', '4')
AND first_channel_source_type not IN ('yqxiu1','yqxiu2','yqxiu3','yqxiu4','yqxiu5','mxyc1','mxyc2','mxyc3'
,'wanpu','jinshan','jx','maimai','zhuoyi','huatian','suopingjingling','mocha','mizhe','meika','lamabang'
,'js-az1','js-az2','js-az3','js-az4','js-az5','jfq-az1','jfq-az2','jfq-az3','jfq-az4','jfq-az5','toufang1'
,'toufang2','toufang3','toufang4','toufang5','toufang6','TF-toufang1','TF-toufang2','TF-toufang3','TF-toufang4'
,'TF-toufang5','tf-toufang1','tf-toufang2','tf-toufang3','tf-toufang4','tf-toufang5','benzhan','promotion_aso100'
,'promotion_qianka','promotion_xiaoyu','promotion_dianru','promotion_malioaso','promotion_malioaso-shequ'
,'promotion_shike','promotion_julang_jl03','promotion_zuimei')
AND first_channel_source_type not LIKE 'promotion\\_jf\\_%') as t2
ON t1.cl_id = t2.device_id
AND t1.partition_date = t2.partition_date
LEFT JOIN
(
SELECT DISTINCT device_id
FROM ml.ml_d_ct_dv_devicespam_d --去除机构刷单设备，即作弊设备（浏览和曝光事件去除）
WHERE partition_day='{}'
UNION ALL
SELECT DISTINCT device_id
FROM dim.dim_device_user_staff --去除内网用户
)spam_pv
on spam_pv.device_id=t1.cl_id
LEFT JOIN
(
SELECT partition_date,device_id
FROM
(--找出user_id当天活跃的第一个设备id
SELECT user_id,partition_date,
if(size(device_list) > 0, device_list [ 0 ], '') AS device_id
FROM online.ml_user_updates
WHERE partition_date>='{}' AND partition_date<'{}'
)t1
JOIN
( --医生账号
SELECT distinct user_id
FROM online.tl_hdfs_doctor_view
WHERE partition_date = '{}'
--马甲账号/模特用户
UNION ALL
SELECT user_id
FROM ml.ml_c_ct_ui_user_dimen_d
WHERE partition_day = '{}'
AND (is_puppet = 'true' or is_classifyuser = 'true')
UNION ALL
--公司内网覆盖用户
select distinct user_id
from dim.dim_device_user_staff
UNION ALL
--登陆过医生设备
SELECT distinct t1.user_id
FROM
(
SELECT user_id, v.device_id as device_id
FROM online.ml_user_history_detail
LATERAL VIEW EXPLODE(device_history_list) v AS device_id
WHERE partition_date = '{}'
)t1
JOIN
(
SELECT device_id
FROM online.ml_device_history_detail
WHERE partition_date = '{}'
AND is_login_doctor = '1'
)t2
ON t1.device_id = t2.device_id
)t2
on t1.user_id=t2.user_id
group by partition_date,device_id
)dev
on t1.partition_date=dev.partition_date and t1.cl_id=dev.device_id
WHERE (spam_pv.device_id IS NULL or spam_pv.device_id ='')
and (dev.device_id is null or dev.device_id ='')
""".format(start, end, reg, start, end, end, start, end, end, end, end, end)
# print("sql", flush=True)
# print(sql, flush=True)
df = spark.sql(sql)
return df
def get_device_click_tractate_ids_dict(click_df):
res = defaultdict(list)
cols = click_df.orderBy("partition_date", ascending=False).collect()
for i in cols:
card_id = i["card_id"]
session_id = i["app_session_id"]
if card_id not in res[session_id]:
res[session_id].append(card_id)
return res
def save_clicked_tractate_ids_item2vec():
click_ids = []
with open(os.path.join(DATA_PATH, "click_tractate_ids.csv"), "r") as f:
data = f.readlines()
for i in data:
tmp = i.split("|")
# app_session_id = tmp[0]
ids = tmp[1].rstrip("\n").split(",")
click_ids.append(ids)
model = Word2Vec(click_ids, hs=0, min_count=3, workers=multiprocessing.cpu_count(), iter=10)
print(model)
print(len(click_ids))
model.save(tractate_click_ids_model_path)
return model
if __name__ == "__main__":
begin_time = time.time()
spark = get_spark("tractate_click_ids")
click_df = get_tracate_click_data(spark, get_ndays_before_no_minus(180), get_ndays_before_no_minus(1))
click_df.show(5, False)
res_dict = get_device_click_tractate_ids_dict(click_df)
with open(os.path.join(DATA_PATH, "click_tractate_ids.csv"), "w") as f:
for (k, v) in res_dict.items():
if v:
f.write("{}|{}\n".format(k, ",".join([str(x) for x in v])))
print("write data done.")
save_clicked_tractate_ids_item2vec()
for id in ["84375", "148764", "368399"]:
print(TRACTATE_CLICK_IDS_MODEL.wv.most_similar(id, topn=5))
print("total cost: {:.2f}mins".format((time.time() - begin_time) / 60))
# spark-submit --master yarn --deploy-mode client --queue root.strategy --driver-memory 16g --executor-memory 1g --executor-cores 1 --num-executors 70 --conf spark.default.parallelism=100 --conf spark.storage.memoryFraction=0.5 --conf spark.shuffle.memoryFraction=0.3 --conf spark.locality.wait=0 --jars /srv/apps/tispark-core-2.1-SNAPSHOT-jar-with-dependencies.jar,/srv/apps/spark-connector_2.11-1.9.0-rc2.jar,/srv/apps/mysql-connector-java-5.1.38.jar /srv/apps/strategy_embedding/word_vector/tractate.py