none
Multi-Dimensional Forecasting with Time Series SQL 2008 RRS feed

  • Question

  • Hello,

    In addition to time, I have 2 dimensions (keys) used to identify data in my model. I need to forecast based on all 3 of these dimensions using the Time Series algorithm in SQL Server 2k8. I am wondering how to model this. Naturally, I will use TIME as the Time Key and one of the other dimensions as the KEY column. Any advice on using the other dimension?

    Would I create a VALUE/PREDICT column for each item in that other dimension? There could be 100s of items:

    TIME  DIM1 DIM2_ITEM1 DIM2_ITEM2 DIM2_ITEM3 DIM2_ITEM4 ...

    OR, would I create a compound dimension for the KEY column, joining my 2 non-TIME dimensions:

    TIME DIM1-DIM2 VALUE

    Thanks in advance, and my apologies if my question seems to obtuse. I would be happy to clarify if needed.

    -- EngFM
    Tuesday, June 2, 2009 1:50 AM

All replies

  • I have the same issue.  Curious if there are any ideas.
    Tuesday, June 23, 2009 3:06 AM
  • You will need to create a compound key for all the non-time dimensions.   You can either architect your cube this way, or take the result of an MDX query and feed that into a mining model directly.
    -- Jamie MacLennan -- SQL Server Data Mining -- http://blogs.msdn.com/jamiemac
    Tuesday, June 23, 2009 6:21 PM