Window based averages based on ranges within another table - snowflake-cloud-data-platform

Question

I have a table that has positions like so:
create or replace table data (
pos int not null,
val float not null,
constraint data_pk primary key (pos)
);
And, a ranges type table like so:
create or replace table ranges (
label varchar(32) not null,
left int not null,
right int not null,
constraint ranges_pk primary key (label)
);
with ranges like
('alpha', 11, 13),
('bravo', 11, 14),
('charlie', 11, 15),
('echo', 12, 18),
('delta', 12, 19),
('foxtrot', 13, 20)
for each label, I need to lookup every possible 3 subrange within the "data" table, take these 3 subrange averages, and then average them...
I couldn't think of a good way to describe what I'm after, so I thought I'd show what I'd expect for 'charlie':
The results for charlie in the select should be:
('charlie', 40.111), -- avg(avg(data[pos=11], data[pos=12], data[pos=13]), avg(data[pos=12], data[pos=13], data[pos=14]), avg(data[pos=13], data[pos=14], data[pos=15]))
-- -> avg(avg(31, 37, 41), avg(37, 41, 43), avg(41, 43, 47))
-- -> avg(36.333, 40.333, 43.667) -> 40.111
(for data like)
insert into data (pos, val) values
(1, 2), (2, 3), (3, 5), (4, 7), (5, 11), (6, 13), (7, 17), (8, 19),
(9, 23), (10, 29), (11, 31), (12, 37), (13, 41), (14, 43), (15, 47), (16, 53),
(17, 59), (18, 61), (19, 67), (20, 71), (21, 73), (22, 79), (23, 83), (24, 89),
(25, 97), (26, 101), (27, 103), (28, 107), (29, 109), (30, 113), (31, 127), (32, 131),
(33, 137), (34, 139), (35, 149), (36, 151), (37, 157), (38, 163), (39, 167), (40, 173),
(41, 179), (42, 181), (43, 191), (44, 193), (45, 197), (46, 199), (47, 211), (48, 223),
(49, 227), (50, 229), (51, 233), (52, 239), (53, 241), (54, 251);
Is there a way to do this within Snowflake SQL? Or must I resort to python to do this? If it helps I made a gist with more data..
Thanks!

Answer 1

Is there a way to do this within Snowflake SQL? Or must I resort to python to do this?
SQL language is expressive enough to handle such case.
Key point here is to use windowed average with windows size of 3 and then average moving averages:
WITH cte AS (
SELECT r.label, r.left, r.right, d.val,
AVG(d.val) OVER(PARTITION BY r.label ORDER BY d.pos ROWS
BETWEEN 2 PRECEDING AND CURRENT ROW) AS r
FROM ranges r
JOIN data d
ON d.pos BETWEEN r.left AND r.right
QUALIFY ROW_NUMBER() OVER(PARTITION BY r.label ORDER BY d.pos) > 2
)
SELECT label, AVG(r) AS output
FROM cte
GROUP BY label
ORDER BY label;
Output:
Intermediate step to ilustrate:
WITH cte AS (
SELECT r.label, r.left, r.right, d.val,
AVG(d.val) OVER(PARTITION BY r.label ORDER BY d.pos
ROWS BETWEEN 2 PRECEDING AND CURRENT ROW) AS r
FROM ranges r
JOIN data d
ON d.pos BETWEEN r.left AND r.right
QUALIFY ROW_NUMBER() OVER(PARTITION BY r.label ORDER BY d.pos) > 2
)
SELECT *
FROM cte
ORDER BY label, r;
Output:

Answer 2

Here is a step by step answer:
WITH data(pos, val) AS (
SELECT * FROM VALUES
(1, 2), (2, 3), (3, 5), (4, 7), (5, 11), (6, 13), (7, 17), (8, 19),
(9, 23), (10, 29), (11, 31), (12, 37), (13, 41), (14, 43), (15, 47), (16, 53),
(17, 59), (18, 61), (19, 67), (20, 71), (21, 73), (22, 79), (23, 83), (24, 89),
(25, 97), (26, 101), (27, 103), (28, 107), (29, 109), (30, 113), (31, 127), (32, 131),
(33, 137), (34, 139), (35, 149), (36, 151), (37, 157), (38, 163), (39, 167), (40, 173),
(41, 179), (42, 181), (43, 191), (44, 193), (45, 197), (46, 199), (47, 211), (48, 223),
(49, 227), (50, 229), (51, 233), (52, 239), (53, 241), (54, 251)
), codes(name,s_val, e_val) AS (
SELECT * FROM VALUES
('alpha', 11, 13),
('bravo', 11, 14),
('charlie', 11, 15),
('echo', 12, 18),
('delta', 12, 19),
('foxtrot', 13, 20)
), ranges as (
SELECT row_number() over (order by null)-1 as seq
FROM table(generator(rowcount => 200))
), boost_codes AS (
select c.name
,c.s_val + r.seq + 0 as b1
,c.s_val + r.seq + 2 as b3
from codes as c
join ranges as r
ON r.seq <= (e_val - s_val - 2)
), almost_there AS (
select
bc.name
,avg(d.val) as partial
from boost_codes as bc
join data as d ON d.pos between bc.b1 and bc.b3
GROUP BY 1, bc.b1
)
SELECT name
,round(avg(partial),3) as output
FROM almost_there
GROUP BY 1
ORDER BY 1;
which gives:
NAME
OUTPUT
alpha
36.333
bravo
38.333
charlie
40.111
delta
50.778
echo
48.467
foxtrot
55.111