Is there a way to count the number of unique value across multiple columns in SQL

Question

I would like to count the number of unique values based on the tx_id , here is part of the raw data:

table : Treatment Record 
+------------------+-----------+----------------+------------------+
|        SN        |  tx_id    |       pa3      |       pa4        |
+------------------+-----------+----------------+------------------+
| I2120210007014   |   149362  | V16F2021117016 |   V15S2021145018 |
| I2120210007014   |   149362  | V15S2021144019 |   V15S2021145018 |
| I2120210007014   |   149362  | V16F2021117017 |   V15S2021145018 |
| I2120210007014   |   149362  | V16F2021117017 |   V15S2021145018 |
| I2120210007014   |   149362  | V16F2021117017 |   V15S2021145018 |
| I2120210007014   |   148716  | V15C2021116010 |   V15C20211091016|
+------------------+-----------+----------------+------------------+

For example, the result should look like below:

+------------------+-----------+----------------+-------+
|        SN        |  tx_id    |  V16F |  V15S  |  V15C |
+------------------+-----------+-------+--------+-------+
| I2120210007014   |   149362  |   2   |    2   |   0   |
| I2120210007014   |   148716  |   0   |    0   |   2   |
+------------------+-----------+----------------+-------+

From the raw data, you can see that there are two different tx_id and I use this to identify each of the group. Therefore, for insance, all the tx_id = '149362 is in the same group.

And in the column of pa3 and pa4 , there are 2 different group which can be categorized by looking at the first 4 characters, like "V16F", "V15S". Moreover, I have to count the number of different wordings in the same group. For example, you can see that column pa3 contains V16F2021117016 , V15S2021144019 , V16F2021117017 while column pa4 contains only . Therefore, there are V15S2021145018 .

Therefore, we count 2 for the group "V16F" and 2 for group "V15S". You may notice that the counting is not based on the columns pa3 and pa4 but based on the last 4 characters. For example V16F2021117016 and V16F2021117017 , they belong to the same group,"V16F", but different words since the last 4 characters are '7016' and '7017' respectively.

However I could not find a way out at this moment and only typed some sql code in below. Hopes someone can help me.

SELECT tx_id, 
       sum(case when val like 'V16F%' then 1 else 0 end), 
       sum(case when val2 like 'V15S%' then 1 else 0 end) 
FROM ( select tx_id, pa3 as val, pa4 as val2 from Cool group by pa3, pa4)
GROUP BY tx_id   

Here is the wrong output:

+------------------+-----------+----------------+
|        SN        |  tx_id    |  V16F |  V15S  |
+------------------+-----------+-------+--------+
| I2120210007014   |   149362  |   3   |    3   |
| I2120210007014   |   148716  |   0   |    0   |  
+------------------+-----------+----------------+

Answer 1

The simplest way to do it is to use UNION ALL to get all pa3 s and pa4 s in 1 column and then aggregate:

SELECT SN, tx_id,
       COUNT(DISTINCT CASE WHEN pa LIKE 'V16F%' THEN pa END) V16F,
       COUNT(DISTINCT CASE WHEN pa LIKE 'V15S%' THEN pa END) V15S,
       COUNT(DISTINCT CASE WHEN pa LIKE 'V15C%' THEN pa END) V15C
FROM (
  SELECT SN, tx_id, pa3 pa FROM tablename
  UNION ALL
  SELECT SN, tx_id, pa4 pa FROM tablename
) t  
GROUP BY SN, tx_id

Or, with UNION , which removes duplicate rows, so there is no need for DISTINCT :

SELECT SN, tx_id,
       COUNT(CASE WHEN pa LIKE 'V16F%' THEN pa END) V16F,
       COUNT(CASE WHEN pa LIKE 'V15S%' THEN pa END) V15S,
       COUNT(CASE WHEN pa LIKE 'V15C%' THEN pa END) V15C
FROM (
  SELECT SN, tx_id, pa3 pa FROM tablename
  UNION 
  SELECT SN, tx_id, pa4 pa FROM tablename
) t  
GROUP BY SN, tx_id

Which can be further simplified to:

SELECT SN, tx_id,
       SUM(pa LIKE 'V16F%') V16F,
       SUM(pa LIKE 'V15S%') V15S,
       SUM(pa LIKE 'V15C%') V15C
FROM (
  SELECT SN, tx_id, pa3 pa FROM tablename
  UNION 
  SELECT SN, tx_id, pa4 pa FROM tablename
) t  
GROUP BY SN, tx_id

Another way, is to directly use conditional aggregation with more complicated logic that works for this sample data:

SELECT SN, tx_id,
       COUNT(DISTINCT CASE WHEN pa3 LIKE 'V16F%' THEN pa3 END) +
       COUNT(DISTINCT CASE WHEN pa4 LIKE 'V16F%' THEN pa4 END) -
       SUM(pa3 = pa4) V16F,
       COUNT(DISTINCT CASE WHEN pa3 LIKE 'V15S%' THEN pa3 END) +
       COUNT(DISTINCT CASE WHEN pa4 LIKE 'V15S%' THEN pa4 END) -
       SUM(pa3 = pa4) V15S,
       COUNT(DISTINCT CASE WHEN pa3 LIKE 'V15C%' THEN pa3 END) +
       COUNT(DISTINCT CASE WHEN pa4 LIKE 'V15C%' THEN pa4 END) -
       SUM(pa3 = pa4) V15C
FROM tablename
GROUP BY SN, tx_id

See the demo .