枫下：我是渣渣，我是想……enm先查缺用pandas或者sql，看看空值原因处理，用可视化初步查异常，箱线图看极端，直方看分布，然后视情况清洗。如果已有特征标签，少就加维度，多的减维度，找效用好的特征。然后回归预测未来，分类进行用户划分，聚类寻找未知特征。

snakajima：第二题暴力解 select uid from user_info where gender='女' and age<='25' and uid in (select uid from user_video where date>='20190801' and date<='20190807' and tag=='美食' group by uid having count(uid)>10) and uid in (select uid from user_video where date=='20190808' and tag=='美食' group by uid having count(uid)<3) 第三题 select distinct tag, max(cnt) over (partition by tag)-min(cnt) over (partition by tag) from (select tag, uid, count(uid) cnt from user_video group by tag, uid) 第一题求大佬 突然发现个问题，他所有的字段都是varchar的……包括年龄这种数字