From a simple table in SQL Server with two columns like below
Key Value
------------
A 5000
B NULL
C 6000
I want to get the first record in the order B,A,C (i.e. get value of B and if null then value of A, and if null, value of C ) where Value is not null. From the above list I expect the output to be 5000
I'm trying with this code - without any luck:
SELECT
CASE
WHEN [Key] = 'B' AND Value IS NOT NULL
THEN Value
WHEN [Key] = 'A' AND Value IS NOT NULL
THEN Value
WHEN [Key] = 'C' AND Value IS NOT NULL
THEN Value
END
FROM
temporary
Hmmmm . . . one method uses coalesce():
select coalesce(max(case when [Key] = 'B' then value end),
max(case when [Key] = 'A' then value end),
max(case when [Key] = 'C' then value end)
)
from temporary;
But, I think I would do:
select top 1 t.value
from temporary t
where value is not null and [Key] in ('A', 'B', 'C')
order by charindex([Key], 'B,A,C');
Note that the order by is just a shorthand for getting the preferred ordering. It works for "A", "B", and "C", but might not generalize to all strings.
You can use a where clause to omit null values, order by your custom sorting (using a case expression) and just take the top 1 row:
SELECT TOP 1 value
FROM mytable
WHERE value IS NOT NULL
ORDER BY CASE key WHEN 'B' THEN 0
WHEN 'A' THEN 1
WHEN 'C' THEN 2
ELSE 3
END ASC
Related
Consider the following table
id attribute
1 a
1 a
1 b
2 a
2 a
3 c
4 a
I want to select the ids that have attribute of 'a' only, ie 2 and 4.
Cant select 1 because 1 has 'a' and 'b', cant select 3 because it has 'c' only. We select 2 and 4 because it has 'a' value only.
You can use
SELECT id
FROM YourTable
GROUP BY id
HAVING MAX(attribute) = 'a' AND MIN(attribute) = 'a'
AND COUNT(*) = COUNT(attribute)
the
COUNT(*) = COUNT(attribute)
is to discard any id that have NULL attribute as well as a. Remove this if that is not the semantics you want or the column is not nullable anyway.
Please test this:
SELECT id
FROM attribute
GROUP BY id
HAVING
COUNT(DISTINCT attribute) = 1 AND MIN(attribute)= 'a';
I am stuck with a specific scenario of flattening the data and need help for it. I need the output as flattened data where the column values are not fixed. Due to this I want to restrict the output to fixed set of columns.
Given Table 'test_table'
ID
Name
Property
1
C1
xxx
2
C2
xyz
2
C3
zz
The scenario is, column Name can have any no. of values corresponding to an ID. I need to flatten the data based in such a way that there is one row per ID field. Since the Name field varies with each ID, I want to flatten it for fix 3 columns like Co1, Co2, Co3. The output should look like
ID
Co1
Co1_Property
Co2
Co2_Property
Co3
Co3_Property
1
C1
xxx
null
null
2
C2
xyz
C3
zz
Could not think of a solution using Pivot or aggregation. Any help would be appreciated.
You can use arrays:
select id,
array_agg(name order by name)[safe_ordinal(1)] as name_1,
array_agg(property order by name)[safe_ordinal(1)] as property_1,
array_agg(name order by name)[safe_ordinal(2)] as name_2,
array_agg(property order by name)[safe_ordinal(2)] as property_2,
array_agg(name order by name)[safe_ordinal(3)] as name_3,
array_agg(property order by name)[safe_ordinal(3)] as property_3
from t
group by id;
All current answers are too verbose and involve heavy repetition of same fragments of code again and again and if you need to account more columns you need to copy paste and add more lines which will make it even more verbose!
My preference is to avoid such type of coding and rather use something more generic as in below example
select * from (
select *, row_number() over(partition by id) col
from `project.dataset.table`)
pivot (max(name) as name, max(property) as property for col in (1, 2, 3))
If applied to sample data in your question - output is
If you want to change number of output columns - you just simply modify for col in (1, 2, 3) part of query.
For example if you would wanted to have 5 columns - you would use for col in (1, 2, 3, 4, 5) - that simple!!!
The standard practice is to use conditional aggregation. That is, to use CASE expressions to pick which row goes to which column, then MAX() to collapse multiple rows into individual rows...
SELECT
id,
MAX(CASE WHEN name = 'C1' THEN name END) AS co1,
MAX(CASE WHEN name = 'C1' THEN property END) AS co1_property,
MAX(CASE WHEN name = 'C2' THEN name END) AS co2,
MAX(CASE WHEN name = 'C2' THEN property END) AS co2_property,
MAX(CASE WHEN name = 'C3' THEN name END) AS co3,
MAX(CASE WHEN name = 'C3' THEN property END) AS co3_property
FROM
yourTable
GROUP BY
id
Background info:
Not having an ELSE in the CASE expression implicitly means ELSE NULL
The intention is therefore for each column to recieve NULL from every input row, except for the row being pivoted into that column
Aggregates, such as MAX() essentially skip NULL values
MAX( {NULL,NULL,'xxx',NULL,NULL} ) therefore equals 'xxx'
A similar approach "bunches" the values to the left (so that NULL values always only appears to the right...)
That approach first uses row_number() to give each row a value corresponding to which column you want to put that row in to..
WITH
sorted AS
(
SELECT
*,
ROW_NUMBER() OVER (PARTITION BY id ORDER BY name) AS seq_num
FROM
yourTable
)
SELECT
id,
MAX(CASE WHEN seq_num = 1 THEN name END) AS co1,
MAX(CASE WHEN seq_num = 1 THEN property END) AS co1_property,
MAX(CASE WHEN seq_num = 2 THEN name END) AS co2,
MAX(CASE WHEN seq_num = 2 THEN property END) AS co2_property,
MAX(CASE WHEN seq_num = 3 THEN name END) AS co3,
MAX(CASE WHEN seq_num = 3 THEN property END) AS co3_property
FROM
yourTable
GROUP BY
id
Lets say I have the following table:
FKEY A B C D E F
'A' 1 0 1 0 1 0
'A' 0 1 1 1 0 0
Now i want to make a group by FKEY but I just want to know if the A-F columns has 1 in one, all or none of the grouped rows.. The resulton the above table would be:
FKEY A B C D E F
'A' S S A S S N
..where S is "some", A is "all" and N is "none".
What would be the best approach to make this query. I could so some nested queries, but isnt there a smarter way?
In my real life data, the 1's and 0's are actually DATETIME and NULL's
You can use case and aggregation:
select fkey,
(case when sum(a) = 0 then 'N'
when sum(a) = count(*) then 'A'
else 'S'
end) as a,
(case when sum(b) = 0 then 'N'
when sum(b) = count(*) then 'A'
else 'S'
end) as b,
. . .
from t
group by fkey;
The above assumes that the values are only 0 and 1. If that is the case, you can actually phrase this as:
(case when max(a) = 0 then 'N'
when min(a) = 1 then 'A'
else 'S'
end) as a,
You mentioned that your 0 and 1 are actually null or non null dates. Here's a modified version of Gordon's query that caters for that:
select fkey,
(case when count(datecol) = 0 then 'all dates are null'
when count(datecol) = count(*) then 'all dates are filled'
else 'some are null, some filled'
end) as a,
...
from t
group by fkey;
COUNT(null) is 0, COUNT('2001-01-01') is 1, COUNT(*) is the row count independent of any variable. Hence, if our count of the dates was 0, all must be null. If the count of the dates was equal to the count of the rows, then all must be filled with some value, otherwise it's a mix
I have a SQL query which returns :
id | value
1 a
1 a
1 b
2 a
2 a
I want to get only id who have only the value a. So the id 2
How to do this ?
You can use aggregation and having clause to check if all the rows have value 'a' for a given id:
Using Count:
select id
from t
group by id
having count(*) = count(case when value = 'a' then 1 end);
Or using Sum
select id
from t
group by id
having SUM(case when value = 'a' then 0 else 1 end) = 0;
Use the next code:-
Select id
from #test
group by id
having sum (case when value = 'a' then 0 else 1 end) = 0
The clue is passing 0 for 'a' and pass 1 for other, then having sum equals 0
This is slightly slower than #Gurwinder Singh's answer but can be more readable if performance is not your top priority.
CREATE TABLE tmp (id int, [value] char(1))
INSERT INTO tmp values (1,'a'),(1,'a'),(1,'b'),(2,'a'),(2,'a')
SELECT DISTINCT id
FROM tmp a
WHERE [value] = 'a'
AND id NOT IN (
SELECT id FROM tmp
WHERE [value] <> 'a')
I have sql like:
select avg(decode(type, 'A', value, null) + decode(type, 'B', value, null)) from table;
The problem with this is some of these types can be null, so the addition part will result in null because adding anything to null makes it null. So you might think I could change the decode from null to 0, but that seems to make the avg() count it as part of it's averaging, but it shouldn't/I don't want it counted as part of the average.
Ideally the addition would just ignore the nulls and just not try to add them to the rest of the values.
So let's say my numbers are:
5 + 6 + 5
3 + 2 + 1
4 + null + 2
They total 28 and I'd want to divide by 8 (ignore the null), but if I change the null to 0 in the decode, the avg will then divide by 9 which isn't what I want.
As written, your code should always return null, since if the first decode returns value, then the second decode must always return null. I'm going to assume that you made an error in genericizing your code and that what you really meant was this:
avg(decode(type1, 'A', value1, null) + decode(type2, 'B', value2, null))
(Or, instead of type1, it could be a.type. The point is that the fields in the two decodes are meant to be separate fields)
In this case, I think the easisest thing to do is check for nulls first:
avg(case when type1 is null and type2 is null then null
else case type1 when 'A' then value1 else 0 end
+ case type2 when 'B' then value2 else 0 end
end)
(I replaced decode with case because I find it easier to read, but, in this case decode would work just as well.)
This is overcomplicated to do a sum here. Juste output the values with a CASE, and you are done.
SELECT AVG(
CASE WHEN type = 'A' OR type = 'B'
THEN value
ELSE null
END
)
FROM table
A simple workaround would be to calculate the average yourself:
select
-- The sum of all values with type 'A' or 'B'
sum(decode(type, 'A', value, 'B', value, 0)) /
-- ... divided by the "count" of all values with type 'A' or 'B'
sum(decode(type, 'A', 1, 'B', 1, 0))
from table;
A SQLFiddle example
But the way AVG() works, it would probably be sufficient, if you just removed the addition and put everything in a single DECODE()
select avg(decode(type, 'A', value, 'B', value, null)) from table
The logic here is a bit complicated:
select avg((case when type = 'A' then value else 0 end) + (case when type = 'B' then value else 0 end))
from table
where type in ('A', 'B')
The where clause guarantees that you have at least one "A" or "B". The problem is arising when you have no examples of "A" or "B".