I have a theoretical question, so I'm not interested in alternative solutions. Sorry.
Q: Is it possible to get the window running function values for all previous rows, except current?
For example:
with
t(i,x,y) as (
values
(1,1,1),(2,1,3),(3,1,2),
(4,2,4),(5,2,2),(6,2,8)
)
select
t.*,
sum(y) over (partition by x order by i) - y as sum,
max(y) over (partition by x order by i) as max,
count(*) filter (where y > 2) over (partition by x order by i) as cnt
from
t;
Actual result is
i | x | y | sum | max | cnt
---+---+---+-----+-----+-----
1 | 1 | 1 | 0 | 1 | 0
2 | 1 | 3 | 1 | 3 | 1
3 | 1 | 2 | 4 | 3 | 1
4 | 2 | 4 | 0 | 4 | 1
5 | 2 | 2 | 4 | 4 | 1
6 | 2 | 8 | 6 | 8 | 2
(6 rows)
I want to have max and cnt columns behavior like sum column, so, result should be:
i | x | y | sum | max | cnt
---+---+---+-----+-----+-----
1 | 1 | 1 | 0 | | 0
2 | 1 | 3 | 1 | 1 | 0
3 | 1 | 2 | 4 | 3 | 1
4 | 2 | 4 | 0 | | 0
5 | 2 | 2 | 4 | 4 | 1
6 | 2 | 8 | 6 | 4 | 1
(6 rows)
It can be achieved using simple subquery like
select t.*, lag(y,1) over (partition by x order by i) as yy from t
but is it possible using only window function syntax, without subqueries?
Yes, you can. This does the trick:
with
t(i,x,y) as (
values
(1,1,1),(2,1,3),(3,1,2),
(4,2,4),(5,2,2),(6,2,8)
)
select
t.*,
sum(y) over w as sum,
max(y) over w as max,
count(*) filter (where y > 2) over w as cnt
from t
window w as (partition by x order by i
rows between unbounded preceding and 1 preceding);
The frame_clause selects just those rows from the window frame that you are interested in.
Note that in the sum column you'll get null rather than 0 because of the frame clause: the first row in the frame has no row before it. You can coalesce() this away if needed.
SQLFiddle
Related
Having the following fields in a table...
+---------+---+---+
| myTime | x | y |
+---------+---+---+
| 13:00 | 0 | 0 |
| 13:05 | 2 | 1 |
| 13:10 | 4 | 2 |
| 13:15 | 1 | 3 |
+---------+---+---+
I need to generate a third one (z) as follows...
+---------+---+---+---+
| myTime | x | y | z |
+---------+---+---+---+
| 13:00 | 0 | 0 | 0 |
| 13:05 | 2 | 1 | 1 |
| 13:10 | 4 | 2 | 3 |
| 13:15 | 1 | 3 | 1 |
+---------+---+---+---+
In the first row z will have a value of 0 and in the next ones, z will be calculated as x-y + (previous row's) z.
I've tried using the row number for each record and LAG to try reading values from previous rows...
WITH rows_sorted AS
(SELECT *, ROW_NUMBER() OVER (ORDER BY myTime) AS row_num
FROM table)
SELECT myTime, x, y
IF(row_num = 1, 0, x - y + LAG(z, 1) OVER (ORDER BY row_num)) AS z
FROM rows_sorted
ORDER BY row_num
...but evidently wouldn't work as in LAG(z, 1), z has not been generated yet.
Any suggestion on how such a thing can be done? I'm using standard SQL in Google BigQuery
Thanks in advance
Since the text above oversimplifies the real calculation, here's a closer approach to what I need to achieve:
+---------+----+----+----+
| myTime | x | y | z |
+---------+----+----+----+
| 13:00 | 15 | 22 | 0 |
| 13:05 | 7 | 21 | 0 |
| 13:10 | 7 | 5 | 2 |
| 13:15 | 9 | 16 | 0 |
| 13:20 | 14 | 5 | 9 |
+---------+----+----+----+
Where z for each row is calculated as follows:
WHEN row_number() = 1 THEN z = 0 (already achieved thanks to the
answer below)
WHEN x+(previous row's)z < y THEN z = 0
WHEN x+(previous row's)z >= y THEN z = x+(previous row's)z - y
Hmmm . . . You can get what you want using:
select t.*,
sum(x - y) over (order by mytime) as z
from t;
The first row has values of 0 for all the columns, so this works for your sample data. If you wanted to explicitly set it to 0, then:
select t.*,
(case when row_number() over order by mytime) = 1
then 0
else sum(x - y) over (order by mytime) - first_value(x - y) over (order by mytime)
end) as z
from t;
This subtracts out the value from the first row from the cumulative sum. However, that seems unnecessary.
Here's some sample data from my table:
day_number daily_users_count
1 1
3 1
6 1
7 1
9 2
10 2
I need all day_number values, from 1 to max(day_number), and I want daily_users_count to be zero if it isn't mentioned in this table.
It should look something like this:
day_number daily_users_count
1 1
2 0
3 1
4 0
5 0
6 1
7 1
8 0
9 2
10 2
I think a left join with a table which has a number column with all integers from 1 to max(day_number) would work, if I put a default value for daily_users_count as 0.
What I don't get is how to create such a table where all integers within a certain range are present. Any alternate solutions or any ways to do this would be much appreciated.
You can do it with a recursive CTE which will return all the day_numbers including the missing ones and then a LEFT join to the table:
with cte as (
select min(day_number) day_number from tablename
union all
select day_number + 1 from cte
where day_number < (select max(day_number) from tablename)
)
select c.day_number,
coalesce(t.daily_users_count, 0) daily_users_count
from cte c left join tablename t
on t.day_number = c.day_number
See the demo.
Results:
| day_number | daily_users_count |
| ---------- | ----------------- |
| 1 | 1 |
| 2 | 0 |
| 3 | 1 |
| 4 | 0 |
| 5 | 0 |
| 6 | 1 |
| 7 | 1 |
| 8 | 0 |
| 9 | 2 |
| 10 | 2 |
I've got a table in a Redshift database that contains intervals which are grouped and that potentially overlap, like so:
| interval_id | l | u | group |
| ----------- | -- | -- | ----- |
| 1 | 1 | 10 | A |
| 2 | 2 | 5 | A |
| 3 | 5 | 15 | A |
| 4 | 26 | 30 | B |
| 5 | 28 | 35 | B |
| 6 | 30 | 31 | B |
| 7 | 44 | 45 | B |
| 8 | 56 | 58 | C |
What I would like to do is to determine the length of the union of the intervals within group. That is, for each interval take u - l, sum over all group members and then subtract off the length of the overlaps between the intervals.
Desired result:
| group | length |
| ----- | ------ |
| A | 14 |
| B | 10 |
| C | 2 |
This question has been asked before, alas it seems that all of the solutions in that thread use features that Redshift doesn't support.
This is not difficult but requires multiple steps. The key is to define the "islands" within each group and then aggregate over those. Lots of subquerys, aggregations, and window functions.
select groupId, sum(ul)
from (select groupId, (max(u) - min(l) + 1) as ul
from (select t.*,
sum(case when prev_max_u < l then 1 else 0 end) over (order by l) as grp
from (select t.*,
max(u) over (order by l rows between unbounded preceding and 1 preceding) as prev_max_u
from t
) t
) t
group by groupid, grp
) g
group by groupId;
The idea is to determine if there is an overlap at the beginning of each record. For this purpose, it uses a cumulative max function on all preceding records. Then, it determines if there is an overlap by comparing the previous max with the current l -- a cumulative sum of overlaps defines a group.
The rest is just aggregation. And more aggregation.
Given a table like this in PostgreSQL:
Messages
message_id | creating_user_id | receiving_user_id | created_utc
-----------+------------------+-------------------+-------------
1 | 1 | 2 | 1424816011
2 | 3 | 2 | 1424816012
3 | 3 | 2 | 1424816013
4 | 1 | 3 | 1424816014
5 | 1 | 3 | 1424816015
6 | 2 | 1 | 1424816016
7 | 2 | 1 | 1424816017
8 | 1 | 2 | 1424816018
I want to get the newest two rows per creating_user_id/receiving_user_id where the other user_id is 1. So the result of the query should look like:
message_id | creating_user_id | receiving_user_id | created_utc
-----------+------------------+-------------------+-------------
1 | 1 | 2 | 1424816011
4 | 1 | 3 | 1424816014
5 | 1 | 3 | 1424816015
6 | 2 | 1 | 1424816016
Using a window function with row_number() I can get the first 2 messages for each creating_user_id or the first 2 messages for each receiving_user_id, but I'm not sure how to get the first two messages for per creating_user_id/receiving_user_id.
Since you filter rows where one of both columns is 1 (and irrelevant), and 1 happens to be the smallest number of all, you can simply use GREATEST(creating_user_id, receiving_user_id) to distill the relevant number to PARTITION BY. (Else you could employ CASE.)
The rest is standard procedure: calculate a row number in a subquery and select the first two in the outer query:
SELECT message_id, creating_user_id, receiving_user_id, created_utc
FROM (
SELECT *
, row_number() OVER (PARTITION BY GREATEST (creating_user_id
, receiving_user_id)
ORDER BY created_utc) AS rn
FROM messages
WHERE 1 IN (creating_user_id, receiving_user_id)
) sub
WHERE rn < 3
ORDER BY created_utc;
Exactly your result.
SQL Fiddle.
My objective is to make dynamic group of lines (of product by TYPE & COLOR in fact)
I don't know if it's possible just with one select query.
But : I want to create group of lines (A PRODUCT is a TYPE and a COLOR) as per the number_per_group column and I want to do this grouping depending on the date order (Order By DATE)
A single product with a NB_PER_GROUP number 2 is exclude from the final result.
Table :
-----------------------------------------------
NUM | TYPE | COLOR | NB_PER_GROUP | DATE
-----------------------------------------------
0 | 1 | 1 | 2 | ...
1 | 1 | 1 | 2 |
2 | 1 | 2 | 2 |
3 | 1 | 2 | 2 |
4 | 1 | 1 | 2 |
5 | 1 | 1 | 2 |
6 | 4 | 1 | 3 |
7 | 1 | 1 | 2 |
8 | 4 | 1 | 3 |
9 | 4 | 1 | 3 |
10 | 5 | 1 | 2 |
Results :
------------------------
GROUP_NUMBER | NUM |
------------------------
0 | 0 |
0 | 1 |
~~~~~~~~~~~~~~~~~~~~~~~~
1 | 2 |
1 | 3 |
~~~~~~~~~~~~~~~~~~~~~~~~
2 | 4 |
2 | 5 |
~~~~~~~~~~~~~~~~~~~~~~~~
3 | 6 |
3 | 8 |
3 | 9 |
If you have another way to solve this problem, I will accept it.
What about something like this?
select max(gn.group_number) group_number, ip.num
from products ip
join (
select date, type, color, row_number() over (order by date) - 1 group_number
from (
select op.num, op.type, op.color, op.nb_per_group, op.date, (row_number() over (partition by op.type, op.color order by op.date) - 1) % nb_per_group group_order
from products op
) sq
where sq.group_order = 0
) gn
on ip.type = gn.type
and ip.color = gn.color
and ip.date >= gn.date
group by ip.num
order by group_number, ip.num
This may only work if your nb_per_group values are the same for each combination of type and color. It may also require unique dates, but that could probably be worked around if required.
The innermost subquery partitions the rows by type and color, orders them by date, then calculates the row numbers modulo nb_per_group; this forms a 0-based count for the group that resets to 0 each time nb_per_group is exceeded.
The next-level subquery finds all of the 0 values we mapped in the lower subquery and assigns group numbers to them.
Finally, the outermost query ties each row in the products table to a group number, calculated as the highest group number that split off before this product's date.