如何订购 sql 中不同字符串的计数

Question

So I have a table which looks like this:

ID           timestamp                  USER_TYPE
osdfbouwefo  2021-03-04 15:58:02.271    unidentified
ieqiofbeoww  2021-03-04 19:58:53.125    GroupA
fbruwbfewef  2021-03-04 20:59:02.273    GroupA
oewfbewfuff  2021-03-05 04:34:12.512    GroupB
hmithneregn  2021-03-05 15:43:22.271    GroupA
v_eifb3r39f  2021-03-06 03:58:54.231    unidentified
ieqiofbeoww  2021-03-06 12:21:34.211    GroupA
fbruwbfewef  2021-03-06 18:56:44.121    GroupA

What I need to create is a running tally of unique users by USER_TYPE. That is, the first time a user appears on the table the are counted for that day, and never counted again. Additionally it needs to be broken down by USER_TYPE and what I expect is a grouped table by DATE accounting for each new user ID, increasing in count as the dates increase for each USER_TYPE.

Final outcome:

DATE           USER_TYPE USE_COUNT
2021-03-04  unidentified         1
2021-03-04        GroupA         2
2021-03-05        GroupB         1
2021-03-05        GroupA         3
2021-03-06  unidentified         2

So if you look at just one USER_TYPE, it increases based on the last count. There is one unidentified in 2021-03-04 so it is represented as 1 in USE_COUNT. The next time unidentified appears with a distinct ID is on 2021-03-06 making it the second time it has appeared, thus this is 2. Same goes for all USER_TYPE's, they are always adding onto themselves.

Notice the final two entries in the original table are not included because those ID's already occurred

Here's what I tried but this isn't exactly it, I hope this is possible!!!

    SELECT
        DISTINCT DATE(TIMESTAMP) AS "DATE",
        USER_TYPE,
        COUNT(ID) OVER (
            PARTITION BY USER_TYPE
            ORDER BY
                DATE(TIMESTAMP) ASC
        ) AS USE_COUNT
    FROM
        table
    ORDER BY
        DATE(TIMESTAMP) ASC

Answer 1

I think this will work:

SELECT DATE(timestamp) DATE,USER_TYPE, 
       ROW_NUMBER() OVER (PARTITION BY USER_TYPE ORDER BY timestamp) USE_COUNT FROM
 (SELECT ID,timestamp,user_type, 
         ROW_NUMBER() OVER (PARTITION BY id ORDER BY timestamp) rnum
 FROM mytable) A
 WHERE rnum=1
 ORDER BY DATE(timestamp) ASC, USER_TYPE DESC;

The idea is first to assign ROW_NUMBER() partitioned by ID and order by timestamp . Then turn it into a sub-query. In the outer query, do another ROW_NUMBER() but this time partition it by USER_TYPE with same ordering as in the sub-query. Based on your sample data, the result should return like this:

DATE	USER_TYPE	USE_COUNT
2021-03-04	unidentified	1
2021-03-04	GroupA	1
2021-03-04	GroupA	2
2021-03-05	GroupB	1
2021-03-05	GroupA	3
2021-03-06	unidentified	2

And here's a fiddle: https://dbfiddle.uk/?rdbms=mysql_8.0&fiddle=3fe5824ea1010d33777a005041d31bda

Answer 2

The accepted answer isn't even close to your expected result set.

You need to find the earliest date per user, either

FROM
 (
   SELECT
      ID
     ,timestamp
     ,user_type
     ,MIN(timestamp)
      OVER (PARTITION BY id) AS min_ts
   FROM mytable
 ) AS dt
WHERE timestamp = min_ts

or

FROM
 (
   SELECT
      ID
     ,timestamp
     ,user_type
     ,ROW_NUMBER() -- min timestamp gets lowest rownum 1
      OVER (PARTITION BY id
            ORDER BY timestamp) AS rn
   FROM mytable
 ) AS dt
WHERE rn=1

Then you count the unique users per day and run a cumulative sum:

SELECT
   CAST(timestamp AS DATE) AS DATE
  ,USER_TYPE
  ,SUM(COUNT(*)) -- cumulative sum over count
       OVER (PARTITION BY USER_TYPE
             ORDER BY CAST(timestamp AS DATE)) AS USE_COUNT 
FROM
 (
   SELECT
      ID
     ,timestamp
     ,user_type
     ,ROW_NUMBER()
      OVER (PARTITION BY id
            ORDER BY timestamp) AS rn
   FROM mytable
 ) AS dt
WHERE rn=1
GROUP BY CAST(timestamp AS DATE), USER_TYPE
ORDER BY DATE, USER_TYPE
;

See fiddle

Answer 3

Hmmm. . . I think you want a cumulative sum window function with aggregation:

select date(timestamp), user_type,
       sum(count(*)) over (partition by user_type order by date(timestamp)) as running_count
from t
group by date(timestamp), user_type;