如何在 Bigquery 的不同值范围内查找表中行的聚合值？

Question

我有一个 bigquery 表，格式为 company_id、date、sales_amount。 sales_amount 是一个 FLOAT64 列，其值可以从 0 到 10 亿不等。我需要为每个 company_id 找到第一次达到特定 sales_amount 范围的第一个日期。

到目前为止，我所写的是每个范围都使用了 with 子句：

With A as (
SELECT company_id, min(date) breakDate
FROM <table>
WHERE sales_amount >= 100000 and sales_amount < 500000
GROUP BY company_id
),
B as (
SELECT company_id, min(date) breakDate
FROM <table>
WHERE sales_amount >= 500000 and sales_amount < 1000000
GROUP BY company_id
),
AllUnion AS (
SELECT * FROM A
LEFT JOIN B
USING(company_id)
WHERE B.breakDate > A.breakDate OR B.company_id is NULL

UNION ALL
SELECT * FROM B
)

因此，当添加新范围时，我必须添加一个新的 With 部分，并在最后一个大的 union 部分合并所有中断事件。在合并时，我将确保如果先发生高阶事件，则过滤掉低阶事件。例如，在这种情况下，一家公司在 1 月份（第一次）的销售额超过 500K，而他们的销售额在 2 月份下降并达到 120K。只有 500K 的销售额事件将被返回，2 月的事件将被过滤掉

我必须为不同的表执行此操作，并且可能有更多事件，我想知道是否有一种聪明的方法可以在 bigquery 中编写此查询？

Answer 1

您可以对桶中的sales_amount 进行分桶，使其具有相同的桶 id。然后通过company_id和bucket_id分组，就可以得到每个bucket的MIN(date)。

• https://cloud.google.com/bigquery/docs/reference/standard-sql/mathematical_functions#range_bucket

SELECT company_id, MIN(date) AS breakDate 
  FROM <table>
 WHERE sales_amount >= 100000 
 GROUP BY company_id, RANGE_BUCKET(sales_amount, [100000, 500000, 1000000]);

例子：

WITH sales AS (
  SELECT 'c1' AS company_id, '2022-05-01' AS date, 99999 AS sales_amount
   UNION ALL
  SELECT 'c1' AS company_id, '2022-05-02' AS date, 100000 AS sales_amount
   UNION ALL
  SELECT 'c1' AS company_id, '2022-05-03' AS date, 499999 AS sales_amount
   UNION ALL
  SELECT 'c1' AS company_id, '2022-05-04' AS date, 500000 AS sales_amount
   UNION ALL
  SELECT 'c1' AS company_id, '2022-05-05' AS date, 1100000 AS sales_amount
)  
SELECT company_id, 
       buckets[SAFE_OFFSET(RANGE_BUCKET(sales_amount, buckets) - 1)] AS bucket_id,
       MIN(sales_amount) AS sales_amount,
       MIN(date) AS breakDate
  FROM sales, UNNEST([STRUCT([100000, 500000, 1000000] AS buckets)])
 WHERE sales_amount >= 100000 
 GROUP BY company_id, bucket_id
;

输出：

示例 2：

WITH sales AS (
  SELECT 'c1' AS company_id, '2022-05-01' AS date, 99999 AS sales_amount
   UNION ALL
  SELECT 'c1' AS company_id, '2022-05-02' AS date, 100000 AS sales_amount
   UNION ALL
  SELECT 'c1' AS company_id, '2022-05-03' AS date, 499999 AS sales_amount
   UNION ALL
  SELECT 'c1' AS company_id, '2022-05-05' AS date, 1100000 AS sales_amount
   UNION ALL
  SELECT 'c1' AS company_id, '2022-05-07' AS date, 500000 AS sales_amount
),
bucketized_sales AS (
  SELECT company_id, 
         buckets[SAFE_OFFSET(RANGE_BUCKET(sales_amount, buckets) - 1)] AS bucket_id,
         MIN(sales_amount) AS sales_amount,
         MIN(date) AS breakDate
    FROM sales, UNNEST([STRUCT([100000, 500000, 1000000] AS buckets)])
   WHERE sales_amount >= 100000 
   GROUP BY company_id, bucket_id
)
SELECT *, MAX(bucket_id) OVER (PARTITION BY company_id ORDER BY breakDate ROWS  BETWEEN  UNBOUNDED PRECEDING AND 1 PRECEDING)  factor
  FROM bucketized_sales 
 WHERE TRUE QUALIFY bucket_id > factor OR factor IS NULL
 ORDER BY breakDate
;

输出：