BigQuery SPLIT() and grouping by result - google-bigquery

Question

Using SPLIT() & NTH(), I'm splitting a string value, and taking the 2nd substring as the result. I then want to group on that result. However, when I use SPLIT() in conjunction with a GROUP BY, it keeps giving the error:
Error: (L1:55): Cannot group by an aggregate
The result is a string, so why is it not possible to group on it?
For example, this works and returns the correct string:
SELECT NTH(2,SPLIT('FIRST-SECOND','-')) as second_part FROM [FOO.bar] limit 10
But then grouping on the result does not work:
SELECT NTH(2,SPLIT('FIRST-SECOND','-')) as second_part FROM [FOO.bar] GROUP BY second_part limit 10

Answer 1

My best guess would be that you can get an equivalent result by using a subquery. Something like :
SELECT * FROM (Select NTH(2,SPLIT('FIRST-SECOND','-')) as second_part FROM [FOO.bar] limit 10) GROUP BY second_part
The system returns Nth in an aggregate internally I guess

Answer 2

If there are always just 2 values separated by a delimeter, then a simpler approach would be to use REGEXP_EXTRACT:
SELECT REGEXP_EXTRACT('FIRST-SECOND','-(.*)') as second_part
from [FOO.bar]
GROUP BY second_part
limit 10

Answer 3

I like David's answer - sometimes splitting can get a bit more complicated using RegEx. Extracting the first option from a split command, then GROUPing BY is a very common operation. The way I normally do this in BigQuery is using a REGEXP_EXTRACT as follows:
In this simple example, the column "splitme" is pipe-delimited (|).
SELECT REGEXP_EXTRACT(splitme, r'(?U)^(.*)\|') AS title, COUNT(*) as c
FROM [my_table]
GROUP BY title;
This means, extract the string from the beginning of "splitme" to the first occurrence of a pipe (|). The "(?U)" is the "un-greedy" match flag in the re2 RegEx engine's syntax. Without this flag, if there are multiple pipe-delimited values, this RegEx would match everything up until the last pipe.

Answer 4

In my practice, I am usually using something like below with N being number of values in "list" to skip.
SELECT REGEXP_EXTRACT(string + '|', r'(?U)^(?:.*\|){N}(.*)\|') AS substring
So if I would be interested in third value in list I would use:
SELECT
REGEXP_EXTRACT(string + '|', r'(?U)^(?:.*\|){2}(.*)\|') AS substring,
COUNT(1) AS weight
FROM yourtable
GROUP BY 1
More details on re2 syntax here