window recipe not behaving as per explanation in concept video

piyushk
piyushk Dataiku DSS Core Designer, Dataiku DSS & SQL, Dataiku DSS Adv Designer, Registered Posts: 55 ✭✭✭✭✭

I am undergoing Advanced Designer Training and wanted to understand Window recipe. (Visual Recipes 101 -> Window Recipe -> Concept)

As per the video, if no window frame is selected, aggregation will be calculated on the partition (as whole). But, this is not happening in Window Recipe for Dataiku DSS v9.0.3.

Screenshots:

1. customer_orders

customer_orders.png

2. window recipe configuration (average over amt)

window_configuration.png

3. resulting dataset

customer_orders_windows.png


Expected output as per video:

expected_output.png

Even enabling the order does not give the desired output. I sincerely think, window recipe is not behaving as expected. Please help me with the case, whether it is a genuine bug or error with the instance.

I was able to reproduce the expected output with following configuration.

Screenshots:

1. window recipe configuration

window_configuration_2.png

2. resulting dataset

customer_orders_windows_expected.png

I have observed the similar case with Hands-On: Window Recipe (Visual Recipes 102 -> Window Recipe (Advanced)). Here the final output is not matching with the one, mentioned in Hands-On section.

Can you help me with the reference project to match datasets with the project mentioned in Hands-On: Window Recipe of (Visual Recipes 102 -> Window Recipe (Advanced)). I was able to match datasets using "Hands-On: Advanced Formula & Regex".

With purchase_date ordered as descending, it gives inconsistent result. Can you please help with it?

Answers

  • Manuel
    Manuel Alpha Tester, Dataiker Alumni, Dataiku DSS Core Designer, Dataiku DSS ML Practitioner, Dataiku DSS Core Concepts, Dataiku DSS Adv Designer, Registered Posts: 193 ✭✭✭✭✭✭✭

    Hi,

    I was able to replicate what you describe, but I believe it is the video that is wrong, not the windows recipe:

    - When only specifying the partition, the aggregation works on the partition value as a whole, but only on the rows seen so far, in the order they are stored.

    - You can test this by changing the first date to 2021, you can see that row will continue to be the first date for Carol.

    Thanks for reporting this issue, I will pass to the Academy team.

  • piyushk
    piyushk Dataiku DSS Core Designer, Dataiku DSS & SQL, Dataiku DSS Adv Designer, Registered Posts: 55 ✭✭✭✭✭

    I have observed the similar case with Hands-On: Window Recipe (Visual Recipes 102 -> Window Recipe (Advanced)). Here the final output is not matching with the one, mentioned in Hands-On section.

    Can you help me with the reference project to match datasets with the project mentioned in Hands-On: Window Recipe of (Visual Recipes 102 -> Window Recipe (Advanced)). I was able to match datasets using "Hands-On: Advanced Formula & Regex".

    With purchase_date ordered as descending, it gives inconsistent result in comparison to purchase_date ordered as ascending (equivalent window frame was selected). Can you please help with it?

  • NancyK
    NancyK Dataiker, Dataiku DSS Core Designer, Dataiku DSS Adv Designer Posts: 22 Dataiker

    Hi @piyushk
    and @Manuel
    , thank you for bringing this up, we have logged the issue and are working on fixing it. I'll respond in this thread to let you know as soon as it's been fixed.

    Best,

    Nancy from the Academy team

Setup Info
    Tags
      Help me…