甚至用PostgreSQL进行数据采样

时间:2016-03-21 03:55:30

标签: sql postgresql

我有一个查询返回两个时间戳之间的所有点。如果我做一个特别大的时间片(比如1年),我可能会得到10000行。我希望能够要求一个分辨率(比如说1天)并让它们均匀间隔1天,并收到~365行。以下是我现在的查询:

SELECT *
      FROM checkins
      WHERE serial=${serial} AND created_at BETWEEN ${startTimestamp} AND ${endTimestamp}
      ORDER BY created_at DESC
      LIMIT ${limit}
      OFFSET ${offset}

关于使用Postgres的好策略的任何想法?

1 个答案:

答案 0 :(得分:0)

假设您有 PG 9.4 + ,这应该可以解决问题:

SELECT *
FROM checkins
JOIN (
  -- The below returns 366 created_at values within the two time points, inclusive
  SELECT precentile_disc(fraction/365.) WITHIN GROUP (ORDER BY created_at) 
  FROM checkins, generate_series(0, 365) f(fraction)
  WHERE serial = ${serial} AND created_at BETWEEN ${startTimestamp} AND ${endTimestamp}
) USING (created_at)
ORDER BY created_at DESC;

percentile_disc() function根据指定的分数从排序组中为您提供离散值,其值最接近提供的分数。结合generate_series(),您可以在分数[0., 0.004, 0.008, ..., 1.]处获得一系列此类值。然后将这些值(created_at值,而不是分数)加回到checkins表中,以获得最终结果。

对于PG的旧版本,您可以手动执行此操作"像这样:

SELECT *
FROM (
  SELECT *, rank() OVER (ORDER BY created_at) AS rnk
  FROM checkins
  WHERE serial = ${serial} AND created_at BETWEEN ${startTimestamp} AND ${endTimestamp}
) sub
WHERE rnk % extract(day from ${endTimestamp} - ${startTimestamp}) = 1
ORDER BY created_at;

这为startTimestampendTimestamp之间的每一天提供了一行,因此,如果它们相隔一年就会获得365行。