Trying to transpose a table in GBQ. The solution needs to be able to work dynamically with varying columns and rows. I know there is a way to do this by listing out columns specifically, but I have varying tables with large numbers of rows. There is an unpivot function in GBQ now and my idea was to Unpivot, then Pivot it. Not sure if anyone has dealt with this yet?
Starting Data
Result
Consider below approach
select * from (
select * from data
unpivot (value for Fruit in (Apple, Bananas, Orange))
)
pivot (max(value) for `Group` in ('North', 'South', 'West'))
If applied to sample data in your question - output is
Below version builds all references needed for unpivot and pivot dynamically on fly
execute immediate (select '''select * from (
select * from `project.dataset.table`
unpivot (value for Fruit in (''' || (select array_to_string(regexp_extract_all(to_json_string((select as struct * except(`Group`) from unnest([t]))), r'"([^"]+)":'), ', ')
from `project.dataset.table` t limit 1) || '''))
)
pivot (max(value) for `Group` in (''' || (select '"' || string_agg(distinct `Group`, '", "' order by `Group`) || '"' from `project.dataset.table`) || '''))
'''
);
Related
I have table with one column called cols. I want to separate cols to new column based on type: SKU and MARK using bigquery.
cols
"dsc":[{"amount":30000,"c_amount":0,"d_amount":0,"d_id":"","scope":"CART","title":"Promo","type":"SKU"},{"amount":7000,"c_amount":0,"d_amount":7000,"d_id":"x","scope":"CART_D","title":"","type":"MARK"}]
The result i want is like:
sku_amount sku_c_amount sku_d_amount sku_d_id sku_scope sku_title mark_amount mark_c_amount mark_d_amount mark_d_id mark_scope
30000 0 0 CART Promo 7000 0 0 x CART_D
Anyone know the script? thank you
Consider below approach
create temp table temp_table as
select id1,
max(if(key = 'type', value, null)) over (partition by id1, id2) || '_' || key as key,
value, offset
from (
select md5(cols) as id1, md5(json) id2, arr[offset(0)] as key, arr[offset(1)] as value, offset
from your_table, unnest(json_extract_array('{' || cols || '}', '$.dsc')) json with offset,
unnest(split(translate(json, '{}"', ''))) kv,
unnest([struct(split(kv, ':') as arr)])
);
execute immediate (select '''
select * except(id1) from (select * except(offset) from temp_table)
pivot (any_value(value) for key in ("''' || string_agg(key, '","' order by offset, key) || '''"))
'''
from (select distinct key, offset from temp_table where not ends_with(key, '_type'))
);
if applied to sample data in your question - output is
I am building an application/script for users that do not have write access to the database. Normally I would use Execute Immediate and save that result into a table, and then pull from that table and continue on with the script. Is there a way to save results from Execute Immediate in either a sub query or CTE so that the script can continue on?
Example code where results are put into a table
execute immediate (select '''create or replace table `project.dataset.table2` as
select `Group`, ''' || (select string_agg('cast(' || Fruit || ' as float64) as ' || Fruit ) from (
select regexp_extract_all(to_json_string((select as struct * except(`Group`) from unnest([t]))), r'"([^"]+)":') Fruits
from `project.dataset.table` t limit 1), unnest(Fruits) Fruit) ||
''' from `project.dataset.table`''')
;
I would need something more along the lines of this, but it doesn't work
WITH CTEtable2 as (
execute immediate (select `Group`, ''' || (select string_agg('cast(' || Fruit || ' as float64) as ' || Fruit ) from (
select regexp_extract_all(to_json_string((select as struct * except(`Group`) from unnest([t]))), r'"([^"]+)":') Fruits
from `project.dataset.table` t limit 1), unnest(Fruits) Fruit) ||
''' from `project.dataset.table`''')
)
SELECT *
FROM CTEtable2
If customer first_name-'Monika',
last_name='Awasthi'
Then I am using below query to return value in json format:
SELECT *
FROM
(
SELECT JSON_ARRAYAGG(JSON_OBJECT('CODE' IS '1','VALUE' IS 'Monika'||' '||'Awasthi'))
FROM DUAL
);
It is working fine & give below output:
[{"CODE":"1","VALUE":"Monika Awasthi"}]
But I want one more value which should be reversed means output should be:
[{"CODE":"1","VALUE":"Monika Awasthi"},{"CODE":"2","VALUE":"Awasthi Monika"}]
Kindly give me some suggestions. Thank You
Another approach is to use a CTE to generate the two codes and values; your original version could be written to get the name data from a table or CTE:
-- CTE for sample data
WITH cte (first_name, last_name) AS (
SELECT 'Monika', 'Awasthi' FROM DUAL
)
-- query against CTE or table
SELECT JSON_ARRAYAGG(JSON_OBJECT('CODE' IS '1','VALUE' IS last_name ||' '|| first_name))
FROM cte;
And you could then extend that with a CTE that generates the value with the names in both orders:
WITH cte1 (first_name, last_name) AS (
SELECT 'Monika', 'Awasthi' FROM DUAL
),
cte2 (code, value) AS (
SELECT 1 AS code, first_name || ' ' || last_name FROM cte1
UNION ALL
SELECT 2 AS code, last_name || ' ' || first_name FROM cte1
)
SELECT JSON_ARRAYAGG(JSON_OBJECT('CODE' IS code,'VALUE' IS value))
FROM cte2;
which gives:
JSON_ARRAYAGG(JSON_OBJECT('CODE'ISCODE,'VALUE'ISVALUE))
-------------------------------------------------------------------------
[{"CODE":1,"VALUE":"Monika Awasthi"},{"CODE":2,"VALUE":"Awasthi Monika"}]
db<>fiddle
A simple logic through use of SQL(without using PL/SQL) in order to generate code values as only be usable for two columns as in this case might be
SELECT JSON_ARRAYAGG(
JSON_OBJECT('CODE' IS tt.column_id,
'VALUE' IS CASE WHEN column_id=1
THEN name||' '||surname
ELSE surname||' '||name
END)
) AS result
FROM t
CROSS JOIN (SELECT column_id FROM user_tab_cols WHERE table_name = 'T') tt
where t is a table which hold name and surname columns
Demo
More resilient solution might be provided through use of PL/SQL, even more columns exist within the data source such as
DECLARE
v_jso VARCHAR2(4000);
v_arr OWA.VC_ARR;
v_arr_t JSON_ARRAY_T := JSON_ARRAY_T();
BEGIN
FOR c IN ( SELECT column_id FROM user_tab_cols WHERE table_name = 'T' )
LOOP
SELECT 'JSON_OBJECT( ''CODE'' IS '||MAX(c.column_id)||',
''VALUE'' IS '||LISTAGG(column_name,'||'' ''||')
WITHIN GROUP (ORDER BY ABS(column_id-c.column_id))
||' )'
INTO v_arr(c.column_id)
FROM ( SELECT * FROM user_tab_cols WHERE table_name = 'T' );
EXECUTE IMMEDIATE 'SELECT '||v_arr(c.column_id)||' FROM t' INTO v_jso;
v_arr_t.APPEND(JSON_OBJECT_T(v_jso));
END LOOP;
DBMS_OUTPUT.PUT_LINE(v_arr_t.STRINGIFY);
END;
/
Demo
As I explained in a comment under your question, I am not clear on how you define the CODE values for your JSON string (assuming you have more than one customer).
Other than that, if you need to create a JSON array of objects from individual strings (as in your attempt), you probably need to use JSON_ARRAY rather than JSON_ARRAYAGG. Something like I show below. Incidentally, I also don't know why you needed to SELECT * FROM (subquery) - the outer SELECT seems entirely unnecessary.
So, if you don't actually aggregate over a table, but just need to build a JSON array from individual pieces:
select json_array
(
json_object('CODE' is '1', 'VALUE' is first_name || ' ' || last_name ),
json_object('CODE' is '2', 'VALUE' is last_name || ' ' || first_name)
) as result
from ( select 'Monika' as first_name, 'Awasthi' as last_name from dual )
;
RESULT
------------------------------------------------------------------------------
[{"CODE":"1","VALUE":"Monika Awasthi"},{"CODE":"2","VALUE":"Awasthi Monika"}]
I would like to calculate, for each column in a table, the percent of rows that are null.
For one column, I was using:
SELECT ((SELECT COUNT(Col1)
FROM Table1)
/
(SELECT COUNT(*)
FROM Table1)) AS Table1Stats
Works great and is fast.
However, I want to do this for all ~50 columns of the table, and my environment does not allow me to use dynamic SQL.
Any recommendations? I am using snowflake to connect to AWS, but as an end user I am using the snowflake browser interface.
You can combine this as:
SELECT COUNT(Col1) * 1.0 / COUNT(*)
FROM Table1;
Or, if you prefer:
SELECT AVG( (Col1 IS NOT NULL)::INT )
FROM Table1;
You can use a mix of object_construct() and flatten() to move the column names into rows. Then do the math for the values missing:
create or replace temp table many_cols as
select 1 a, 2 b, 3 c, 4 d
union all select 1, null, 3, 4
union all select 8, 8, null, null
union all select 8, 8, 7, null
union all select null, null, null, null;
select key column_name
, 1-count(*)/(select count(*) from many_cols) ratio_null
from (
select object_construct(a.*) x
from many_cols a
), lateral flatten(x)
group by key
;
You can do this using a SQL generator if you don't mind copying the text and running it once it's done.
-- SQL generator option:
select 'select' || listagg(' ((select count(' || COLUMN_NAME || ') from "SNOWFLAKE_SAMPLE_DATA"."TPCH_SF10000"."ORDERS") / ' ||
'(select count(*) from "SNOWFLAKE_SAMPLE_DATA"."TPCH_SF10000"."ORDERS")) as ' || COLUMN_NAME, ',') as SQL_STATEMENT
from "SNOWFLAKE_SAMPLE_DATA"."INFORMATION_SCHEMA"."COLUMNS"
where TABLE_CATALOG = 'SNOWFLAKE_SAMPLE_DATA' and TABLE_SCHEMA = 'TPCH_SF10000' and TABLE_NAME = 'ORDERS'
;
If the copy and paste is not plausible because you need to script it, you can use the results of the SQL generator in a stored procedure I wrote to execute a single line of dynamic SQL:
call run_dynamic_sql(
select 'select' || listagg(' ((select count(' || COLUMN_NAME || ') from "SNOWFLAKE_SAMPLE_DATA"."TPCH_SF10000"."ORDERS") / ' ||
'(select count(*) from "SNOWFLAKE_SAMPLE_DATA"."TPCH_SF10000"."ORDERS")) as ' || COLUMN_NAME, ',') as SQL_STATEMENT
from "SNOWFLAKE_SAMPLE_DATA"."INFORMATION_SCHEMA"."COLUMNS"
where TABLE_CATALOG = 'SNOWFLAKE_SAMPLE_DATA' and TABLE_SCHEMA = 'TPCH_SF10000' and TABLE_NAME = 'ORDERS'
);
If you want the stored procedure, until it's published on Snowflake's blog it's available here: https://snowflake.pavlik.us/index.php/2021/01/22/running-dynamic-sql-in-snowflake/
Using PostgreSQL 9.3, I'm creating a Jasper reports template to make a pdf report. I want to create reports of different tables, with multiple columns, all with the same template. A solution could be to get values of register as pairs of column name and value per id.
By example, if I had a table like:
id | Column1 | Column2 | Column3
-------------------------------------------------
1 | Register1C1 | Register1C2 | Register1C3
I would like to get the register as:
Id | ColumnName | Value
-----------------------------
1 | Column1 | Register1C1
1 | Column2 | Register1C2
1 | Column3 | Register1C3
The data type of value columns can vary!
Is it possible? How can I do this?
If all your columns share the same data type and order of rows does not have to be enforced:
SELECT t.id, v.*
FROM tbl t, LATERAL (
VALUES
('col1', col1)
, ('col2', col2)
, ('col3', col3)
-- etc.
) v(col, val);
About LATERAL (requires Postgres 9.3 or later):
What is the difference between LATERAL and a subquery in PostgreSQL?
Combining it with a VALUES expression:
Crosstab transpose query request
SELECT DISTINCT on multiple columns
For varying data types, the common denominator would be text, since every type can be cast to text. Plus, order enforced:
SELECT t.id, v.col, v.val
FROM tbl t, LATERAL (
VALUES
(1, 'col1', col1::text)
, (2, 'col2', col2::text)
, (3, 'col3', col3::text)
-- etc.
) v(rank, col, val)
ORDER BY t.id, v.rank;
In Postgres 9.4 or later use the new unnest() for multiple arrays:
SELECT t.id, v.*
FROM tbl t, unnest('{col1,col2,col3}'::text[]
, ARRAY[col1,col2,col3]) v(col, val);
-- , ARRAY[col1::text,col2::text,col3::text]) v(col, val);
The commented alternative for varying data types.
Full automation for Postgres 9.4:
The query above is convenient to automate for a dynamic set of columns:
CREATE OR REPLACE FUNCTION f_transpose (_tbl regclass, VARIADIC _cols text[])
RETURNS TABLE (id int, col text, val text) AS
$func$
BEGIN
RETURN QUERY EXECUTE format(
'SELECT t.id, v.* FROM %s t, unnest($1, ARRAY[%s]) v'
, _tbl, array_to_string(_cols, '::text,') || '::text'))
-- , _tbl, array_to_string(_cols, ','))) -- simple alternative for only text
USING _cols;
END
$func$ LANGUAGE plpgsql;
Call - with table name and any number of column names, any data types:
SELECT * FROM f_transpose('table_name', 'column1', 'column2', 'column3');
Weakness: the list of column names is not safe against SQL injection. You could gather column names from pg_attribute instead. Example:
How to perform the same aggregation on every column, without listing the columns?
SELECT id
,unnest(string_to_array('col1,col2,col3', ',')) col_name
,unnest(string_to_array(col1 || ',' || col2 || ',' || col3, ',')) val
FROM t
Try following method:
My sample table name is t,to get the n columns name you can use this query
select string_agg(column_name,',') cols from information_schema.columns where
table_name='t' and column_name<>'id'
this query will selects all columns in your table except id column.If you want to specify schema name then use table_schema='your_schema_name' in where clause
To create select query dynamically
SELECT 'select id,unnest(string_to_array(''' || cols || ''','','')) col_name,unnest(string_to_array(' || cols1 || ','','')) val from t'
FROM (
SELECT string_agg(column_name, ',') cols -- here we'll get all the columns in table t
,string_agg(column_name, '||'',''||') cols1
FROM information_schema.columns
WHERE table_name = 't'
AND column_name <> 'id'
) tb;
And using following plpgsql function dynamically creates SELECT id,unnest(string_to_array('....')) col_name,unnest(string_to_array(.., ',')) val FROM t and execute.
CREATE OR replace FUNCTION fn ()
RETURNS TABLE (
id INT
,columname TEXT
,columnvalues TEXT
) AS $$
DECLARE qry TEXT;
BEGIN
SELECT 'select id,unnest(string_to_array(''' || cols || ''','','')) col_name,unnest(string_to_array(' || cols1 || ','','')) val from t'
INTO qry
FROM (
SELECT string_agg(column_name, ',') cols
,string_agg(column_name, '||'',''||') cols1
FROM information_schema.columns
WHERE table_name = 't'
AND column_name <> 'id'
) tb;
RETURN QUERY
EXECUTE format(qry);
END;$$
LANGUAGE plpgsql
Call this function like select * from fn()