Sampling calculation on based IDs in snowflake - snowflake-cloud-data-platform

Question

I've been facing an issue while taking sample from an ID of a table. If I sample it over the whole population then getting improper results, and getting count 3 on CLI_1 and count 0 for CLI_2, when I sample with .1 mean. For example -
Table create query :
CREATE
OR
REPLACE temp TABLE tmp.cg_test AS
SELECT
column1::varchar ID ,
column2::boolean sample_flag ,
column3::varchar term_id
FROM
( VALUES
('ID_1', NULL, 'CLI_1'),
('ID_2', NULL, 'CLI_1'),
('ID_3', NULL, 'CLI_1'),
('ID_4', NULL, 'CLI_1'),
('ID_5', NULL, 'CLI_1'),
('ID_6', NULL, 'CLI_1'),
('ID_7', NULL, 'CLI_1'),
('ID_8', NULL, 'CLI_1'),
('ID_9', NULL, 'CLI_1'),
('ID_10', NULL,'CLI_1'),
('ID_11', NULL,'CLI_2'),
('ID_12', NULL,'CLI_2'),
('ID_13', NULL,'CLI_2'),
('ID_14', NULL,'CLI_2'),
('ID_15', NULL,'CLI_2'),
('ID_16', NULL,'CLI_2'),
('ID_17', NULL,'CLI_2'),
('ID_18', NULL,'CLI_2'),
('ID_19', NULL,'CLI_2'),
('ID_20', NULL,'CLI_2')
);
Sampling Query:
CREATE OR REPLACE TEMP TABLE tmp.sample_calc_cli AS
Select id,
uniform(0::float, 1::float, random(1)) AS random_val,
-- uniform(0::float, 1::float, random(1)) OVER (PARTITION BY term_id) AS cli_sampling_value,
(uniform(0::float, 1::float, random(1))<.1)::boolean AS sample_flag,
-- ((uniform(0::float, 1::float, random(1)) < 0.1) OVER (PARTITION BY term_id))::boolean as sample_flag, -- error
term_id
from tmp.cg_test
GROUP BY id, term_id ORDER BY term_id asc;
Count check for sample:
SELECT term_id,
count(term_id) AS total_cli,
count_if(sample_flag= true) AS sample_count
FROM tmp.sample_calc_cli GROUP BY term_id;
Result
TERM_ID
TOTAL_CLI
SAMPLE_COUNT
CLI_1
10
3
CLI_2
10
0
Uncommenting partitioning with uniform() populates an error as :
SQL Error [2060] [42601]: SQL compilation error:
Invalid function type [UNIFORM] for window function.
What could be the possible way to rectify this issue?

Answer 1

You just calculate random values, to randomly pick rows from the table. If you use 23 as the seed number, you can see that you get 2 samples from CLI_2:
CREATE OR REPLACE TEMP TABLE sample_calc_cli AS
Select id,
uniform(0::float, 1::float, random(23)) AS random_val,
(uniform(0::float, 1::float, random(23))<.1)::boolean AS sample_flag,
term_id
from cg_test
GROUP BY id, term_id ORDER BY term_id asc;
If you need to pick equal number of sample rows, you may use this:
SELECT *
FROM sample_calc_cli
qualify row_number() over (partition by term_id order by random() ) < 3;
This one will pick 2 rows for each term_id.