Unexpected “ghost” values in Window recipe (sum and count not matching)
Hey everyone, im new here
I’m facing a weird issue with the Window recipe while trying to compute rolling transaction metrics — specifically sum and count over 1-day, 7-day, and 30-day periods.
Here’s my setup:
- Partitioned by cc_num
- Ordered by trans_date_trans_time
- Using a window frame (either limit by range in days/hourse or limit by rows i tried all of em)
- I created two separate Window recipes — one for sum and another for count of transactions
The problem is that I’m getting some strange “ghost” accumulation values.
Example:
Let’s say for a single customer I have only two transactions in a day — $5 and $7.
Logically, I expect:
- Transaction 1 → total = $5, count = 1
- Transaction 2 → total = $12, count = 2
But instead, I sometimes get:
- Transaction 1 → total = $5, count = 5 (?!?)
- Transaction 2 → total = $30, count = 7
It’s as if the window is summing or counting transactions that don’t actually exist yet — like it’s pulling data from outside the date range. fyi i use firs record for my sample so its kinda weird
Can anyone help me? Any thoughts, tips, or debugging ideas would be much appreciated 🙏
Answers
-
Hi,
I hope that you are doing well:)
The best approach here would be for us to first see your full window recipes set up and different input and output to best understand the pattern behavior and how it is behaving based on your recipe definition.
Either you can share this information here (if not sensitive) or submit a ticket here: https://support.dataiku.com/support/tickets/new
If you submit a support ticket please do provide the information above as well as the additional information i have requested about your recipe. Sharing the logs of both window recipe run would also be great.
We will be happy to work with you to help troubleshoot your issue!
Best,
Yasmine
