What's the equivalent of Excel's `left(find(), -1)` in BigQuery? - google-bigquery

I have names in my dataset and they include parentheses. But, I am trying to clean up the names to exclude those parentheses.
Example: ABC Company (Somewhere, WY)
What I want to turn it into is: ABC Company
I'm using standard SQL with google big query.
I've done some research and I know big query has left(), but I do not know the equivalent of find(). My plan was to do something that finds the ( and then gives me everything to the left of -1 characters from the (.

My plan was to do something that finds the ( and then gives me everything to the left of -1 characters from the (.
Good plan! In BigQuery Standard SQL - equivalent of LEFT is SUBSTR(value, position[, length]) and equivalent of FIND is STRPOS(value1, value2)
With this in mind your query can look like (which is exactly as you planned)
#standardSQL
WITH names AS (
SELECT 'ABC Company (Somewhere, WY)' AS name
)
SELECT SUBSTR(name, 1, STRPOS(name, '(') - 1) AS clean_name
FROM names
Usually, string functions are less expensive than regular expression functions, so if you have pattern as in your example - you should go with above version
But in more generic cases, when pattern to clean is more dynamic like in Graham's answer - you should go with solution in Graham's answer

Just use REGEXP_REPLACE + TRIM. This will work with all variants (just not nested parentheses):
#standardSQL
WITH
names AS (
SELECT
'ABC Company (Somewhere, WY)' AS name
UNION ALL
SELECT
'(Somewhere, WY) ABC Company' AS name
UNION ALL
SELECT
'ABC (Somewhere, WY) Company' AS name)
SELECT
TRIM(REGEXP_REPLACE(name,r'\(.*?\)',''), ' ') AS cleaned
FROM
names

Use REGEXP_EXTRACT:
SELECT
RTRIM(REGEXP_EXTRACT(names, r'([^(]*)')) AS new_name
FROM yourTable
The regex used here will greedily consume and match everything up until hitting an opening parenthesis. I used RTRIM to remove any unwanted whitespace picked up by the regex.
Note that this approach is robust with respect to the edge case of an address record not having any term with parentheses. In this case, the above query would just return the entire original value.

I can't test this solution at the moment, but you can combine SUBSTR and INSTR. Like this:
SELECT CASE WHEN INSTR(name, '(') > 0 THEN SUBSTR( name, 1, INSTR(name, '(') ) ELSE name END as name FROM table;

Related

How do I dynamically extract substring from string?

I’m trying to dynamically extract a substring from a very long URL. For example, I may have the following URLs:
https://www.google.com/ABCDEF Version=“0.0.00.0” GHIJK
https://www.google.com/ABCDEFGH Version=“0.0.0.0” IJKLM
https://www.google.com/ABC Version=“0.0.0.00” 12345
I am trying to extract the version code only (0.0.0.0).
This is what I have so far:
SELECT SUBSTR(col, INSTR(col, ‘Version=“‘)+9)
FROM table
This query returns the following result:
0.0.00.0” GHIJK … (url continues on)
So, I attempt to find “Version” in the link, so I can start from the same position in each row. This works fine, however I’m having a hard time dynamically locating the ending quote (“). I tried using INSTR in the third parameter of my SUBSTR function, like so:
SELECT SUBSTR(col, INSTR(col, ‘Version=“‘)+9, INSTR(col, ‘“‘))
FROM table
I figured that this would find the position of the ending quote, and then use that number for the length, but it returns a strange output. I’ve also used POSITION, CHARINDEX, LENGTH, and LOCATE. None of these functions work in Oracle.
I think maybe when I put +9 after the first INSTR function, it’s setting the query to a fixed position instead of a dynamic one, but I’m not sure how else to remove ‘Version=“‘.
Here's one option (which, actually, selects what's between double quotes - that's version in your example; if there were some other similar substring, you'd get a wrong result).
with test (col) as
(select 'https://www.google.com/ABCDEF Version="0.0.00.0" GHIJK' from dual union all
select 'https://www.google.com/ABCDEFGH Version="0.0.0.0" IJKLM' from dual union all
select 'https://www.google.com/ABC Version="0.0.0.00" 12345' from dual
)
select col,
replace(regexp_substr(col, '".+"'), '"') version
from test;
which results in
https://www.google.com/ABCDEF Version="0.0.00.0" GHIJK 0.0.00.0
https://www.google.com/ABCDEFGH Version="0.0.0.0" IJKLM 0.0.0.0
https://www.google.com/ABC Version="0.0.0.00" 12345 0.0.0.00
You can still use use INSTR to locate the second " in the string, then subtract the location of the first " to get the length that you need to get. Below is an example query:
SELECT col,
SUBSTR (col, INSTR (col, '"') + 1, INSTR (col, '"', 1, 2) - INSTR (col, '"') - 1) version
FROM test;
You can use REGEXP_SUBSTR() with Version=(\d.*\d?) pattern in order to extract the piece between Version=" and "(your quotes are presumed to be regular double quotes " ")
SELECT REGEXP_SUBSTR(url,'Version="(\d.*\d)"',1,1,null,1) AS version
FROM t
where
the third argument(1) is position,
the fourth argument(1) is occurence, and especially important to use the last one as being capture group (1)
indeed using '"(\d.*\d)"' pattern is enough for the
current data set
or
REGEXP_REPLACE() with capture group \2 as
SELECT REGEXP_REPLACE(url,'^(.*Version=")([^"]*).*','\2') AS version
FROM t
Demo

Regex: how to get the text between a few colons?

So, i have a lot of strings like the ones below in my database:
product1:1stparty:single_aduls:android:
product2:3rdparty:married_adults:ios:
product3:3rdparty:other_adults:android:
I need a regex to get only the text after the product name and before the device category. So, in the first line I'd get 1stparty:single_aduls, in the second 3rdparty:married_adults and in the third 3rdparty:other_adults. I'm stuck and can't find a way to solve that. Could anyone help me please?
As a regular expression, you can use:
select regexp_extract('product1:1stparty:single_aduls:android:', '^[^:]*:(.*):[^:]*:$')
This returns every after the first colon and before the penultimate colon.
We can try using REGEXP_REPLACE here:
SELECT REGEXP_REPLACE(val, r"^.*?:|:[^:]+:$", "") AS output
FROM yourTable;
This approach removes either the leading ...: or trailing :...: from the column, leaving behind the content you want. Here is a demo showing that the regex replacement is working:
Demo
You can also use standard split function and access result array element by index, which is quite clear to read and understand.
with a as (
select split('product1:1stparty:single_aduls:android:', ':') as splitted
)
select splitted[ordinal(2)] || ':' || splitted[ordinal (3)] as subs
from a
Consider below example
with your_table as (
select 'product1:1stparty:single_aduls:android:' txt union all
select 'product2:3rdparty:married_adults:ios:' union all
select 'product3:3rdparty:other_adults:android:'
)
select *,
(
select string_agg(part, ':' order by offset)
from unnest(split(txt, ':')) part with offset
where offset in (1, 2)
) result
from your_table
with output

Replace a recurring word and the character before it

I am using SQL Server trying to replace each recurring "[BACKSPACE]" in a string and the character that came before the word [BACKSPACE] to mimic what a backspace would do.
Here is my current string:
"This is a string that I would like to d[BACKSPACE]correct and see if I could make it %[BACKSPACE] cleaner by removing the word and $[BACKSPACE] character before the backspace."
Here is what I want it to say:
"This is a string that I would like to correct and see if I could make it cleaner by removing the word and character before the backspace."
Let me make this clearer. In the above example string, the $ and % signs were just used as examples of characters that would need to be removed since they are before the [BACKSPACE] word that I want to replace.
Here is another before example:
The dog likq[BACKSPACE]es it's owner
I want to edit it to read:
The dog likes it's owner
One last before example is:
I am frequesn[BACKSPACE][BACKSPACE]nlt[BACKSPACE][BACKSPACE]tly surprised
I want to edit it to read:
I am frequently surprised
Without a CLR function that provides Regex replacement the only way you'll be able to do this is with iteration in T-SQL. Note, however, that the below solution does not give you the results you ask for, but does the logic you ask. You state that you want to remove the string and the character before, but in 2 of your scenarios that isn't true. For the last 2 strings you remove ' %[BACKSPACE]' and ' $[BACKSPACE]' respectively (notice the leading whitespace).
This leading whitespace is left in this solution. I am not entertaining fixing that, as the real solution is don't use T-SQL for this, use something that supports Regex.
I also assume this string is coming from a column in a table, and said table has multiple rows (with a distinct value for the string on each).
Anyway, the solution:
WITH rCTE AS(
SELECT V.YourColumn,
STUFF(V.YourColumn,CHARINDEX('[BACKSPACE]',V.YourColumn)-1,LEN('[BACKSPACE]')+1,'') AS ReplacedColumn,
1 AS Iteration
FROM (VALUES('"This is a string that I would like to d[BACKSPACE]correct and see if I could make it %[BACKSPACE] cleaner by removing the word and $[BACKSPACE] character before the backspace."'))V(YourColumn)
UNION ALL
SELECT r.YourColumn,
STUFF(r.ReplacedColumn,CHARINDEX('[BACKSPACE]',r.ReplacedColumn)-1,LEN('[BACKSPACE]')+1,''),
r.Iteration + 1
FROM rCTE r
WHERE CHARINDEX('[BACKSPACE]',r.ReplacedColumn) > 0)
SELECT TOP (1) WITH TIES
r.YourColumn,
r.ReplacedColumn
FROM rCTE r
ORDER BY ROW_NUMBER() OVER (PARTITION BY r.YourColumn ORDER BY r.Iteration DESC);
dB<>fiddle
I've had a crack to see if I can get this to work using the traditional tally-table method without any recursion.
I think I have something that works - however the recursive cte version is definitely a cleaner solution and probably better performing, however throwing this in as just an alternative non-recursive way.
/* tally table for use below */
select top 1000 N=Identity(int, 1, 1)
into dbo.Digits
from master.dbo.syscolumns a cross join master.dbo.syscolumns
with w as (
select seq = Row_Number() over (order by t.N),
part = Replace(Substring(#string, t.N, CharIndex(Left(#delimiter,1), #string + #delimiter, t.N) - t.N),Stuff(#delimiter,1,1,''),'')
from Digits t
where t.N <= DataLength(#string)+1 and Substring(Left(#delimiter,1) + #string, t.N, 1) = Left(#delimiter,1)
),
p as (
select seq,Iif(Iif(Lead(part) over(order by seq)='' and lag(part) over(order by seq)='',1,0 )=1 ,'', Iif( seq<Max(seq) over() and part !='',Left(part,Len(part)-1),part)) part
from w
)
select result=(
select ''+ part
from p
where part!=''
order by seq
for xml path('')
)
Here's a simple RegEx pattern that should work:
/.\[BACKSPACE\]/g
EDIT
I have no way to test this right now on my chromebook, but this seems like it should work for T-SQL in the LIKE clause
LIKE '_\[BACKSPACE]' ESCAPE '\'

BigQuery return all matches of regular expression

In Big Query, when I do a regular expression search, it only returns the first match/occurrence.
Is there any way to return all matches, concatenated? something like GROUP_CONCAT maybe?
REGEXP_EXTRACT(body, r"(\w+ )")
In Standard SQL that was recently introduced being supported by BigQuery - you can try as below
SELECT
body,
(SELECT STRING_AGG(word) FROM words.word) AS words
FROM (
SELECT
body, REGEXP_EXTRACT_ALL(body, r'(\w+)') AS word
FROM (
SELECT 'abc xyz qwerty asd' AS body UNION ALL
SELECT 'zxc dfg 345' AS body
)
) words
Don't forget to uncheck Use Legacy SQL checkbox under Show Options
See more details on REGEXP_EXTRACT_ALL and STRING_AGG
If you are stuck with what is now in BigQuery called legacy SQL - you can try something like below
SELECT
body,
GROUP_CONCAT(SPLIT(body, ' ')) AS words
FROM
(SELECT 'abc xyz qwerty asd' AS body),
(SELECT 'zxc dfg 345' AS body)
I understand, this is not necessarily exactly what you need - but might help
Another approach with BigQuery legacy SQL that more suited to cases where you have to use regex.
For example - assume you need extract only numbers from body
Idea is to nuke anything but numbers from body using REGEXP_REPLACE and then apply above described SPLIT() + GROUP_CONCAT()
SELECT
body,
GROUP_CONCAT(SPLIT(REGEXP_REPLACE(body, r'(\D)+', ':'), ':')) AS words
FROM
(SELECT 'abc 123 xyz 543 qwerty asd' AS body),
(SELECT '987zxc 123 dfg 345' AS body)

Select statement with column contains '%'

I want to select names from a table where the 'name' column contains '%' anywhere in the value. For example, I want to retrieve the name 'Approval for 20 % discount for parts'.
SELECT NAME FROM TABLE WHERE NAME ... ?
You can use like with escape. The default is a backslash in some databases (but not in Oracle), so:
select name
from table
where name like '%\%%' ESCAPE '\'
This is standard, and works in most databases. The Oracle documentation is here.
Of course, you could also use instr():
where instr(name, '%') > 0
One way to do it is using replace with an empty string and checking to see if the difference in length of the original string and modified string is > 0.
select name
from table
where length(name) - length(replace(name,'%','')) > 0
Make life easy on yourselves and just use REGEXP_LIKE( )!
SQL> with tbl(name) as (
select 'ABC' from dual
union
select 'E%FS' from dual
)
select name
from tbl
where regexp_like(name, '%');
NAME
----
E%FS
SQL>
I read the documentation mentioned by Gordon. The relevent sentence is:
An underscore (_) in the pattern matches exactly one character (as opposed to one byte in a multibyte character set) in the value
Here was my test:
select c
from (
select 'a%be' c
from dual) d
where c like '_%'
The value a%be was returned.
While the suggestions of using instr() or length in the other two answers will lead to the correct answer, they will do so slowly. Filtering on function results simply take longer than filtering on fields.