Question

I need to answer a question like this:

For each user, what is the most items that user viewed in any 60 second
time frame between START_TIMESTAMP and END_TIMESTAMP?

The 60 second time frame is a sliding window. It’s not just a matter of “items viewed” counts for each whole minute. Also, 60 seconds was just an example, it should work for any number of seconds.

My data is stored like this:

-- Timestamped log of users viewing items
CREATE TABLE user_item_views (
  user_id integer,
  item_id integer,
  timestamp timestamp
);

Doing it for each whole minute is easy enough, just format timestamp to something like YYYY-MM-DD hh:mm and do a count grouped by that formatted timestamp and the user_id.

Doing it for a sliding window, I have no idea how to approach.

If this would be easier outside of SQL, I am open to exporting the data to another format, or using another language.

Desired output is something like:

User ID  Max items viewed in N seconds, between START and END.
...      ...
...      ...
...      ...

How can I do this?

Answer 1

Here’s how I would do it (beware, untested code, this ist just to outline the idea).

You need a helper table with as many rows as there are seconds between START_TIMESTAMP and END_TIMESTAMP. Create that as a temp table before you begin your query.

For the sake of the sample, let’s call it every_second. I’m assuming your minimum time resolution is one second.

Then do:

SELECT
  s.timestamp,
  v.user_id,
  (
   SELECT COUNT(*) FROM user_item_views 
    WHERE timestamp BETWEEN s.timestamp AND ADDTIME(s.timestamp, '00:00:59')
          AND user_id = v.user_id
  ) item_count
FROM
  every_second s
  LEFT JOIN user_item_views v ON v.timestamp = s.timestamp
GROUP BY
  s.timestamp, 
  v.user_id

Store that in another temporary table and select the desired maxima from it (this is necessary because of the “select max from group” problem).