time-series prediction INCREDIBLY slow! HELP

Discussion in 'microsoft.public.sqlserver.datamining' started by Elmer Miller, Apr 23, 2007.

  1. Elmer Miller

    Elmer Miller Guest

    I can't believe that the time-series algorithm is really as slow as it
    seems to be for me, so I must be doing something wrong. I'm trying to
    predict 24 weeks of weekly sales. I have created a model with 6 months
    of weekly data. The processing of the model was fast (about 2 minutes)
    so I was initially encouraged, but the prediction queries are WAY too
    slow. It takes over 5 minutes (PER product) to calculate the
    predictions. At this rate it would take 48 days to generate the
    predictions, about since I have about 14000 products! And this is not on
    a slow machine; it is on a three processor server with 12 GB ram.

    Does this performance sound normal? Is time-series prediction not
    intended to be used in this way? Are there any tips anyone knows about
    that could DRASTICALLY improve performance?



    Elmer Miller
     
    Elmer Miller, Apr 23, 2007
    #1
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.