Forum Discussion

Anonymous's avatar
Anonymous
Not applicable
6 years ago
Solved

Star Schema problem

Hi all!

 

I'm struggling a bit with a data model that we are using at work. I'm trying to implement a nice star schema, but I'm running into some restrictions. 

 

It is important to know that all our data is separated by something we call a "program id". All of our facts and dimensions contain this programID, it is what our database is partitioned by.

 

So the problem is the following: 

 

Let's assume I have a transactions table, a store dimension and a program dimension. Star schema rules say that I connect both store and program to the transactions table.


In my visualization, I use a programID slicer coming from the program dimension, which in my real model contains info about the program, like start time and end time. I then display the total sales, and the number of stores in total, regardless of sales, this is where the problem starts.


 

Because of the star schema, the Store table is not filtered by the program table, and the answer, 10, is incorrect (10 is the total number of stores from all programs) . So any metric that I derive from Transaction will work fine, but I will never filter other dimensions by this "master dimension" ProgramID, which is what I need it to do.

 

To Solve this, I can think of 2 ways:

1: Create an inactive relationship between the program table and the store table, and use USERELATIONSHIP. This makes all my metrics quite ugly and harder to work with.

2: Use a variable to determine the MAX of Program[Program ID], and then apply that as a CALCULATE filter for every metric that uses data from Store as its base. Again, Ugly  😞

 

Is there another solution that I'm not thinking of?

Thanks in advance!

Jaap

Need Help
  • v-yingjl's avatar
    v-yingjl
    6 years ago

    Hi Anonymous ï¼Œ

    You are right. Currently userelationship is the better way to solve this issue because your relationships between the store table and transaction table is [store id] instead of [programm id]. In this case, the result of VN06 must be 3 rather than 5 unless you choose userelationship to pass [programm] parameter.

     

    Best Regards,
    Yingjie Li

    If this post helps then please consider Accept it as the solution to help the other members find it more quickly.

9 Replies

  • amitchandak's avatar
    amitchandak
    Super User
    6 years ago

    Anonymous , if program id to store is one to many join them to create a join program , store, fact.

    or Merge program and store

    or try

    calculate(countx(Transactions,related(store[store_id])))

    • Anonymous's avatar
      Anonymous
      Not applicable
      6 years ago

      Hey Amitchandak,

       

      Your first suggestion seems to suggest creating a join between 2 dimensions, but my real data model has about 10 dimensions, of which some are 1,5M rows. Don't think that's a good solution right? Or do I misunderstand?

       

      Same for merging, that works nice in my testcase, but not in a proper data model with many dimensions.

       

      The last option seems less performant than my solutions, because it uses an iterator, no? Also, this still relies on transactions, so any stores that didnt have any transactions would not get counted, which is not what I want. I just want a count of the number of stores in the store table for this particular program.

       

      Thanks for thinking along though!

       

      Jaap