问题聚集度Hive SQL

问题聚集度：最小的分母占比，贡献最多的分子占比，即小规模贡献大问题。
sql 复制代码
select
	city_name
	,user_id
	,rf_type
	,deal_ord_cnt
	,sale_amt
	,rf_ord_cnt
	,rf_amt
	,rf_ra
	,rf_amt_ra
	,rf_all
	,ord_cnt_all
	,rf_gx
	,ord_cnt_gx
	,del_gx
	,row_number() over(partition by rf_type order by del_gx desc,rf_ra desc,user_id) as rn    -- 贡献差由大到小排序
	,sum(rf_ord_cnt) over(partition by rf_type order by del_gx desc,rf_ra desc,user_id) as rf_ordby    -- 分子累计
	,sum(deal_ord_cnt) over(partition by rf_type order by del_gx desc,rf_ra desc,user_id) as ord_cnt_ordby    -- 分母累计
	,(sum(rf_ord_cnt) over(partition by rf_type order by del_gx desc,rf_ra desc,user_id)) / rf_all as rf_ordby_ra    -- 分子累计占比曲线
	,(sum(deal_ord_cnt) over(partition by rf_type order by del_gx desc,rf_ra desc,user_id)) / ord_cnt_all as ord_cnt_ordby_ra    -- 分母累计占比曲线
	,((sum(rf_ord_cnt) over(partition by rf_type order by del_gx desc,rf_ra desc,user_id)) / rf_all -
		(sum(deal_ord_cnt) over(partition by rf_type order by del_gx desc,rf_ra desc,user_id)) / ord_cnt_all) del1    -- 累计贡献差
	,lead(((sum(rf_ord_cnt) over(partition by rf_type order by del_gx desc,rf_ra desc,user_id)) / rf_all -
		(sum(deal_ord_cnt) over(partition by rf_type order by del_gx desc,rf_ra desc,user_id)) / ord_cnt_all),1) 
			over(partition by rf_type order by del_gx desc,rf_ra desc,user_id) - 
			((sum(rf_ord_cnt) over(partition by rf_type order by del_gx desc,rf_ra desc,user_id)) / rf_all -
		(sum(deal_ord_cnt) over(partition by rf_type order by del_gx desc,rf_ra desc,user_id)) / ord_cnt_all) as del2    --二阶差分，大于0的部分为聚集部分
from
(
select
	city_name
	,user_id
	,deal_ord_cnt  -- 分母
	,sale_amt
	,rf_cancel_ord_cnt as rf_ord_cnt    -- 分子
	,rf_qx_amt as rf_amt
	,rf_cancel_ord_cnt / deal_ord_cnt as rf_ra    -- 监控指标
	,rf_qx_amt/sale_amt as rf_amt_ra
	,sum(rf_cancel_ord_cnt) over() as rf_all
	,sum(deal_ord_cnt) over() as ord_cnt_all
	,rf_cancel_ord_cnt / (sum(rf_cancel_ord_cnt) over()) as rf_gx    -- 分子贡献
	,deal_ord_cnt / (sum(deal_ord_cnt) over()) as ord_cnt_gx    -- 分母贡献
	,rf_cancel_ord_cnt / (sum(rf_cancel_ord_cnt) over()) - deal_ord_cnt / (sum(deal_ord_cnt) over()) as del_gx    -- 贡献差值
	,'取消订单' as rf_type
from table_refund
where rf_cancel_ord_cnt > 0    -- 限定分子>0
) t0