CODEWITHSUNDEEP
sqlcountgoogle-bigquery
user3768071
user3768071

Reputation: 887

how to do nested SQL select count

i'm querying a system that won't allow using DISTINCT, so my alternative is to do a GROUP BY to get near to a result

my desired query was meant to look like this,

SELECT 
SUM(column1) AS column1,
SUM(column2) AS column2,
COUNT(DISTINCT(column3)) AS column3
FROM table

for the alternative, i would think i'd need some type of nested query along the lines of this,

SELECT 
SUM(column1) AS column1,
SUM(column2) AS column2,
COUNT(SELECT column FROM table GROUP BY column) AS column3
FROM table

but it didn't work. Am i close?

Upvotes: 3

Views: 40627

Answers (4)

John Bollinger
John Bollinger

Reputation: 180331

You are using the wrong syntax for COUNT(DISTINCT). The DISTINCT part is a keyword, not a function. Based on the docs, this ought to work:

SELECT 
  SUM(column1) AS column1,
  SUM(column2) AS column2,
  COUNT(DISTINCT column3) AS column3
FROM table

Do, however, read the docs. BigQuery's implementation of COUNT(DISTINCT) is a bit unusual, apparently so as to scale better for big data. If you are trying to count a large number of distinct values then you may need to specify a second parameter (and you have an inherent scaling problem).

Update:

If you have a large number of distinct column3 values to count, and you want an exact count, then perhaps you can perform a join instead of putting a subquery in the select list (which BigQuery seems not to permit):

SELECT *
FROM (
    SELECT
      SUM(column1) AS column1,
      SUM(column2) AS column2
    FROM table
  )
  CROSS JOIN (
      SELECT count(*) AS column3
      FROM (
          SELECT column3
          FROM table
          GROUP BY column3
        )
    )

Update 2:

Not that joining two one-row tables would be at all expensive, but @FelipeHoffa got me thinking more about this, and I realized I had missed a simpler solution:

SELECT
  SUM(column1) AS column1,
  SUM(column2) AS column2,
  COUNT(*) AS column3
FROM (
    SELECT
      SUM(column1) AS column1,
      SUM(column2) AS column2
    FROM table
    GROUP BY column3
  )

This one computes a subtotal of column1 and column2 values, grouping by column3, then counts and totals all the subtotal rows. It feels right.

Upvotes: 3

Felipe Hoffa
Felipe Hoffa

Reputation: 59175

Instead of doing a COUNT(DISTINCT), you can get the same results by running a GROUP BY first, and then counting results.

For example, the number of different words that Shakespeare used by year:

SELECT corpus_date, COUNT(word) different_words
FROM (
  SELECT word, corpus_date
  FROM [publicdata:samples.shakespeare]
  GROUP BY word, corpus_date
)
GROUP BY corpus_date
ORDER BY corpus_date

As a bonus, let's add a column that identifies which books were written during each year:

SELECT corpus_date, COUNT(word) different_words, GROUP_CONCAT(UNIQUE(corpus)) books
FROM (
  SELECT word, corpus_date, UNIQUE(corpus) corpus
  FROM [publicdata:samples.shakespeare]
  GROUP BY word, corpus_date
)
GROUP BY corpus_date
ORDER BY corpus_date

Upvotes: 0

JustAPup
JustAPup

Reputation: 1780

You can probably use a CTE

WITH result as (select column from table group by column)

SELECT 
SUM(column1) AS column1,
SUM(column2) AS column2,
Select Count(*) From result AS column3
FROM table

Upvotes: 0

Duffmaster33
Duffmaster33

Reputation: 1190

FWIW, the way you are trying to use DISTINCT isn't how its normally used, as its meant to show unique rows, not unique values for one column in a dataset. GROUP BY is more in line with what I believe you are ultimately trying to accomplish.

Depending upon what you need you could do one of a couple things. Using your second query, you would need to modify your subquery to get a count, not the actual values, like:

SELECT 
SUM(column1) AS column1,
SUM(column2) AS column2,
(SELECT sum(1) FROM table GROUP BY column) AS column3
FROM table

Alternatively, you could do a query off your initial query, something like this:

SELECT sum(column1), sum(column2), sum(column4) from (
SELECT 
SUM(column1) AS column1,
SUM(column2) AS column2,
1 AS column4
FROM table GROUP BY column3)
GROUP BY column4

Edit: The above is generic SQL, not too familiar with Google Big Query

Upvotes: 1

Related Questions