on 08-12-2016 11:05 AM
Hi Guys,
I have the following problem.
Situation:
1. In SAP Lumira Desktop I want to create a linear regression or trend line for dimensions x and y not associated with time (revenue and price).
2. Then I want to calculate the corresponding 'y' for each 'x' (y= alpha + beta*x) that is a point on this trend line. And get the delta/difference between the y of the trend line and the actual y.
What I already did, since the dimensions are not associated with time is create a series of Calculated Measures & Dimensions to get an intercept (alpha) and slope (beta) of the trend line (this works perfectly!). So basically I solved the formula of the trend line (y = alpha + beta*x), with two Measures alpha (1) and beta (2).
Complication:
The Measures alpha and beta are calculated measures, and I cant use - for some reason - the outcome of these Calculated measures to Calculate a new dimension. So I can't calculate the 'y' for the 'x' of each record in the data.
In addition, I can't just use the sum or build on these calculated measures with new calculated measures, since I want to use a filter (which records are under the trend line, and which are above), therefore I need an extra dimension.
Resolution:
My Questions:
1. Is there a way to use these calculated measures for a new dimension?
2. Is there an easier way to get what I want. E.g. another way to calculate the delta/difference?
Thanks a lot in advance for any help!
Rutger
Rutger - my view only here
I don't think Lumira was intended to provide a calculation like you are describing; see this idea on Idea Place Calcualtion Engine in SAP Lumira : View Idea
What you are describing sounds more like a use case for Predictive Analytics which can perform regression analysis - there is a trial available at SAP BusinessObjects Predictive Analytics – Free 30-Day Trial!
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Hi Tammy,
Thanks for your answer!
So what you are saying basically is that you think I cannot solve this problem in Lumira? Or is it not used to solve this problem? I am not familiar with all different SAP software and I am using the basics through a Trial version of Lumira Desktop.
Also, Is this SAP BO Predictive Analysis also predictive for non-time dimensions?
Thanks,
Rutger
Hi Rutger - Lumira is for simple forecasting with some simple predictive features. It is a data discovery tool
I don't know how it would know to take a value of something you've calculated and then re-use it
Predictive Analytics has several features, several algorithms, where you could take the results and reuse them later in your model. There is the Expert Analytics portion where you could likely do what you are describing above (note: I am not a Predictive expert). There are also automated analytics algorithms too.
Tammy, Thanks again.
Unfortunately not the answer I was hoping to find here. I was hoping to get something with a shortcut here. Since a calculated measure is "only" 1 value when you display it as a number in a chart.
But you are right, the measures are calculated using the whole dataset and to apply it per individual record sounds tricky.
Predictive Analytics sounds very cool tho. I will see if I can get that up and running.
Thanks,
Rutger
Hi Rutger,
Would you be able to create the alpha and beta as custom dimensions instead of making it a formula measure (as all measures are also available as dimensions). You can later convert it to a measure if required. Using this approach you can create the dimension "y" using alpha and beta.
Even though these functionalities are not available out of the box, if you have the formulas you can use it as a workaround for creating the desired functionality in Lumira.
Thank you,
Varun Anand
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Hi Varun,
Thanks for your suggestion.
The answer is no, because the alpha and beta are calculated by using multiple records, and dimensions are only calculated by using data from a single record (since it is not possible to use measures right?). I have to calculate the following to get to the alpha and beta:
Measure(8) is now the SLOPE (beta)
Measure(9) is (Measure(8)-Measure(8)*Measure(4))/count is intercept (alpha)
So these are alot workarounds just to get what I want in Lumira, However, in this way the alpha and beta do vary when filters are applied to the dataset, so that is exactly what I need. Now I need to find the difference in value from the records y and the expected y for each x.
Thanks,
Rutger
User | Count |
---|---|
93 | |
10 | |
10 | |
9 | |
9 | |
7 | |
6 | |
5 | |
5 | |
4 |
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.