每 20 次迭代更新 SQL 中的行

Question

我在来自the NYC Yellow TaxiCab public dataset 的 Google BigQuery 中有一个包含约 100 万行的表。正如您从该链接中看到的那样，该模式没有主键。每行代表一次旅行/交易，但没有customer_id 字段。

我想添加一列 customer_id 并向其分配随机数，以便：

For rows 1-20, `customer_id` should be assigned `1`
For rows 21-40, `customer_id` should be assigned `2`
and so on..

换句话说，我希望表中的 20 行恰好（和任何）具有特定值 customer_id。

Answer 1

为每一行分配一个随机 id，为每个 new_id 获取约 20 行的组：

#standardSQL
SELECT CAST(FLOOR(COUNT(*) OVER()/20*RAND()) AS INT64) new_id, *
FROM (
  SELECT login
  FROM `ghtorrent-bq.ght_2017_04_01.users`
  LIMIT 1000000
)

证明超过一百万行时生成了 50,000 个“customers_id”：

Answer 2

以下是 BigQuery 标准 SQL，每个 customer_id 正好生成 20 个条目

#standardSQL
SELECT DIV(ROW_NUMBER() OVER() - 1, 20) AS customer_id, *
FROM `yourTable`
-- ORDER BY customer_id

您可以使用下面的虚拟数据来测试游戏

#standardSQL
WITH `yourTable` AS (
    SELECT login
    FROM `ghtorrent-bq.ght_2017_04_01.users`
    LIMIT 1000000
)
SELECT DIV(ROW_NUMBER() OVER() - 1, 20) AS customer_id, *
FROM `yourTable`
-- ORDER BY customer_id  

此外 - 下面的查询显示了每个 customer_id 的计数分布

#standardSQL
WITH `yourTable` AS (
    SELECT login
    FROM `ghtorrent-bq.ght_2017_04_01.users`
    LIMIT 1000000
)
SELECT cnt, COUNT(1) AS distribution FROM (
  SELECT customer_id, COUNT(1) AS cnt FROM (
    SELECT *, DIV(ROW_NUMBER() OVER() - 1, 20) AS customer_id
    FROM `yourTable`
    ORDER BY customer_id
  )
  GROUP BY customer_id
)
GROUP BY cnt
ORDER BY cnt   

输出如下

Row cnt distribution     
--- --- ------------
1    20        50000