Adding dimensions to the Statistical Product Forecasting APP.

RalfdeGier

Hi everyone, 

I would like to expand the Statistical Product Forecasting App with another set of dimensions. The model now only models the dimension of the products (hierarchy L3) and time. I would like to add another dimension, for example: contract sale and a non contract sale. 

Does anyone have expierence with adding an extra dimension to this model? Is is just simpely adding the list with my extra dimensions to all modules which contain Hierarchy L3 or is it more complex?

Kind regards,
Ralf 

PaulRitner

Hi, 

 

I haven't looked at the statistical app, but I think about the following things when adding a new dimension:

1. Is the new dimension an attribute of an existing dimension?  If the new dimension is an attribute, such as City being an attribute of Store, then I generally create a new "City" module and aggregate the Store data by City.
2. If the new dimension is not an attribute of one of the existing dimensions, then will adding the new dimension (as a separate dimension) create too much sparsity (i.e. will the model fail to scale due to size constraints, or is it taking up more workspace than I'm comfortable with)?  If yes, then I consider adding a new level to my composite hierarchy (for example, possibly adding your new dimension above or under Product).
3. If I am going to add a new dimension, then I usually do the following:
   1. Back up the model
   2. For each module affected
      1. Copy the module
      2. Update the copy
   3. When all the copied modules are updated, I connect them like the existing modules, while at the same time unhooking the existing modules from the model. 
   4. Unit test the build
   5. Delete the old modules. 
   6. I do the work in copied modules so that the formulas don't all have to be unhooked and then reinserted.  Also, the original modules have the "referenced by" column, which helps guide whether the update is complete.

In specific reference to your question:  "Is it just simply adding the list with my extra dimensions to all modules which contain Hierarchy L3 or is it more complex?"  .... Its usually not simple adding a new dimension to an existing module.  Often, the nature of existing formulas breaks down and results in errors (this is primarily why I copy modules above).  In addition, data loads may have to be restaged (usually not difficult, but can take a some time).

 

Cheers!

RalfdeGier

Hi Paul,
Thank you for your comprehensive response. The dimension is not an Attribute, but a whole new dimension. Therefore it will take some time, but I will sure try to implement option (3). It will take some time, but if it would help a lot.

Kind regards,
Ralf

Cam_Haines

Hi Ralf,

 

Not sure if you completed this task but I've done this before so I'll throw in my two cents:

 

1.  Adding another dimension into a model as complex as stat forecasting is time consuming, as you probably know.  When I did it, I exported the entire line items list of the model into excel to identify which line items needed the new dimension added and which line item formulas needed updating to the include that new dimension.  It was tedious and you'll have to be sure everything still works after you're finished.

 

2.  Are you positive you want to add another dimension to the stat forecast?  As a general guideline in statistical forecasting the more granular you get, the less accurate your forecast becomes.  Could you forecast at product and then disaggregate to Product/Customer using a percetage breakdown of history for the last 6 months or year?  Also, adding another dimension to all those stat forecast modules will dramatically increase the size of your model... something to consider.

 

I would suggest option two, unless there are other considerations I'm unaware of.

 

-Cam Haines

 

RalfdeGier

Dear Cam, 

Thank you for your reply.  Like you said in (1), I'm affraid that adding another dimension will make the model unstable. Currently, the development of another model has higher priority. Therefore I will need to pick up this problem later this year. It will be a hard job to do tough. 

Kind regards, 
Ralf 

saibandaru

Hi Paul,

 

While adding this dimension, I saw an issue with Ranking , the Statistical Forecasting "Best method" is chosen based out of RANK which can be applicable to ONLY one set of group(ex: either PRODUCT or either CUSTOMER), now again the issue is with how to RANK based on additional dimensions added to the existing model....

 

Your thoughts on ADDING RANK to 2 or more groups(dimensions)???

 

 

bdeaton

Is a two-step import a possibility here? I could envision a setup where you import the data from your multi-dimensional module to a list or 2D table, rank based on that list, and then import the calculated ranks back into the module where you need the ranks. But that would mean you lose the dynamic aspect of the model since you'd have to manually re-run this process each time you updated the values to be ranked. I don't know if that would mess up any other aspects of the statistical app, however.

RalfdeGier

Dear Sai,
Thank you for your reply. I haven't encountered this problem yet. this is because I currently use a filter which filters the multi dimensional input to a 2dimensional input. Therefore I need to select each item of the '3rd' dimension and save the outcome of the statistical forecast into a savings module. The Rank formula does not have more groups of dimensions in my version of this model (yet). I would like to replace this filter module by adding dimensions to each module. But I haven't got the time and priority yet to do so.

This is an issue to take into account when acutally adding the dimensions to the modules. thanks for the heads up about this possible problem.