get set of random rows with distinct values

• Jon Stranger

  SSC Veteran

  Points: 273

  More actions

  June 17, 2013 at 4:58 am

  #295361

  I have an interesting little T-SQL problem that has me beaten.

  I want to retrieve a small set of randomly selected rows from a table but also ensure that a particular column (AUTHORID) is unique (not duplicated) within that set.

  SELECT TOP 6 WORKID, AUTHORID FROM dbo.LITERARYWORKS

  ORDER BY CHECKSUM(NEWID())

  works fine at getting a random selection, but of course does not prevent duplicates on AUTHORID. And I can't use DISTINCT without including WORKID and AUTHORID in the ORDER BY - and in any case the rows would always be unique if WORKID is included. I believe one way of achieving the effect of a 'DISTINCT' on a single column while returning other columns as well is by using GROUP BY, but I can't see how to use that in combination with ORDER BY.

  Any ideas?

  Jon

• Lowell

  SSC Guru

  Points: 323522

  More actions

  June 17, 2013 at 5:39 am

  #1624549

  by introducing a row_number function that partitions by the two columns, you cna find unique rows; fromt here, it's just the same query you had before, but hitting a subquery isntead of the raw table:

  SELECT

  TOP 6

  WORKID,

  AUTHORID

  FROM (SELECT

  ROW_NUMBER() OVER(PARTITION BY WORKID, AUTHORID ORDER BY WORKID, AUTHORID) AS RW,

  WORKID,

  AUTHORID

  FROM dbo.LITERARYWORKS) MyAlias

  WHERE RW = 1--finds Uniqueness across WORKID, AUTHORID

  ORDER BY

  CHECKSUM(NEWID())

  Lowell

  ---

  --help us help you! If you post a question, make sure you include a CREATE TABLE... statement and INSERT INTO... statement into that table to give the volunteers here representative data. with your description of the problem, we can provide a tested, verifiable solution to your question! asking the question the right way gets you a tested answer the fastest way possible!

• Jon Stranger

  SSC Veteran

  Points: 273

  More actions

  June 17, 2013 at 5:54 am

  #1624551

  Works a treat - many thanks!

• ChrisM@Work

  SSC Guru

  Points: 186127

  More actions

  June 17, 2013 at 6:31 am

  #1624563

  A randomly-selected work from a number of randomly-selected authors:

  SELECT

  WORKID,

  AUTHORID

  FROM (

  SELECT TOP 10

  AUTHORID

  FROM dbo.LITERARYWORKS

  GROUP BY MASTNUM

  ORDER BY NEWID()

  ) at

  CROSS APPLY (

  SELECT TOP 1

  WORKID

  FROM dbo.LITERARYWORKS

  WHERE AUTHORID = at.AUTHORID

  ORDER BY NEWID()

  ) MyAlias

  ^{“Write the query the simplest way. If through testing it becomes clear that the performance is inadequate, consider alternative query forms.” - Gail Shaw}

  For fast, accurate and documented assistance in answering your questions, please read this article.
  Understanding and using APPLY, (I) and (II) Paul White
  Hidden RBAR: Triangular Joins / The "Numbers" or "Tally" Table: What it is and how it replaces a loop Jeff Moden

• Jon Stranger

  SSC Veteran

  Points: 273

  More actions

  June 18, 2013 at 1:21 am

  #1624878

  Many thanks. I actually prefer this for my purposes as it will always return the required number of items (provided that there is at least that number of authors with at least 1 work each) whereas the other approach above is arguably more random but can return less than the required number of items if more than 1 work for the same author is initially selected.