SQL加权平均

Question

如下表所示。

make   | model | engine | cars_checked | avg_mileage
---------------------------------------|--------
suzuki | sx4   | petrol | 11           | 12
suzuki | sx4   | diesel | 150          | 16
suzuki | swift | petrol | 140          | 15
suzuki | swift | diesel | 18           | 19
toyota | prius | petrol | 16           | 17
toyota | prius | hybrid | 250          | 24

所需的输出是

1. 发动机的平均行驶里程（汽油，柴油）
2. 品牌平均行驶里程
3. 车型平均行驶里程

无法做到简单，group by因为cars_checked需要考虑每条记录（）的样本数权重，以避免平均值平均值的问题。

什么是实现它的正确方法？有没有办法考虑样本数量以进行加权平均group by？

更新 -为上面的＃1添加了输出格式作为示例

engine   | mileage_by_engine
--------------------------
petrol   | xx.z
diesel   | yy.z

Answer 1

SELECT engine, SUM(cars_checked * avg_mileage) / SUM(cars_checked) AS avgMilageByEngine
FROM [YOUR_TABLE]
GROUP BY engine

SELECT make, SUM(cars_checked * avg_mileage) / SUM(cars_checked) AS avgMilageByMake
FROM [YOUR_TABLE]
GROUP BY make

SELECT model, SUM(cars_checked * avg_mileage) / SUM(cars_checked) AS avgMilageByModel
FROM [YOUR_TABLE]
GROUP BY model

Answer 2

简化查询的一种方法是使用grouping sets：

select engine, make, model,
       sum(cars_check * avg_mileage) / sum(cars_checked) as avgMilage
from t
group by grouping sets ((engine), (make), (model));

输出格式仅在未聚合的列中具有非 NULL 值。

Answer 3

有更好的方法。创建聚合函数。操作方法如下。

CREATE OR REPLACE FUNCTION public.numeric_weighted_average_accum(
    "Previous" numeric[],
    "ThisDatum" numeric,
    "ThisWeight" numeric)
    RETURNS numeric[]
    LANGUAGE 'sql'
    COST 100
    VOLATILE STRICT PARALLEL UNSAFE
AS $BODY$
 SELECT ARRAY["Previous"[1] + ("ThisDatum" * "ThisWeight"), "Previous"[2] + "ThisWeight"];
$BODY$;

CREATE OR REPLACE FUNCTION numeric_weighted_average_final(
    "NWA" numeric[])
    RETURNS numeric
    LANGUAGE 'sql'
    COST 100
    VOLATILE STRICT PARALLEL UNSAFE
AS $BODY$
SELECT "NWA"[1] / "NWA"[2];
$BODY$;

CREATE OR REPLACE AGGREGATE weighted_average(datum numeric, weight numeric) (
    SFUNC = numeric_weighted_average_accum,
    STYPE = numeric[] ,
    FINALFUNC = numeric_weighted_average_final,
    FINALFUNC_MODIFY = READ_ONLY,
    INITCOND = '{0,0}',
    MFINALFUNC_MODIFY = READ_ONLY
);

然后，你可以做

SELECT name, weighted_average(avgcolumn, weightcolumn) AS "WeightedAverage" GROUP BY name;

我确信这里还有提高效率的空间，并且很高兴听到这些信息。

哈特哈，