3 Replies Latest reply on Dec 31, 2011 3:17 PM by Mark Kelly-Oracle

    installed sd 3.1, can't create output table

      has enyone encountered/resolved this problem? not sure why a view can be created but not a table. event log says 'workflow failed due to unexpected failure in the following nodes, please review if there were any unexpected core dumps: Output_3_5.

      11gr2, win7 64
        • 1. Re: installed sd 3.1, can't create output table
          Mark Kelly-Oracle
          There is a bug creating a table that has as input containing the following 3 apply output column types:

          1) Function: Prediction Bounds Lower
          Param: Confidence (%): 95.0, Target: <Most Likely>
          2) Function: Prediction Bounds Upper
          Param: Confidence (%): 95.0, Target: <Most Likely>
          3) Function: Prediction Probability
          Param: Target: <Most Likely>

          This is fixed in the Oracle Data Mining 11.2.03 release.
          So you can upgrade your db to, or you can avoid having all three columns passed as input to the create table.
          For example, you can use 2 Update Table nodes, each with the same input flow, and just edit the columns that are being updated so these columns are not being inserted at the same time.

          Now if this is not the same issue you are running into, then we will need more information about what input the Create Table node is processing and if you are using any index or key options.
          Thanks, Mark
          • 2. Re: installed sd 3.1, can't create output table
            Thanks, Mark. Table was created after removing the upper and lower limit stats (but retained the SVM prob). Incidentally, the upper and lower limits were not populated.

            does the 11.2.03 rel fix this? how do I check which DM ver is running? how do I upgrade?
            • 3. Re: installed sd 3.1, can't create output table
              Mark Kelly-Oracle
              Regarding upper and lower limits. They can return null depending on the quality of the model build.
              Here is a snippet from the ODM documentation.

              The PREDICTION_BOUNDS function is for use only with generalized linear models. It returns an object with two NUMBER fields LOWER and UPPER. For a regression mining function, the bounds apply to value of the prediction. For a classification mining function, the bounds apply to the probability value. If the GLM was built using ridge regression, or if the covariance matrix is found to be singular during the build, then this function returns NULL for both fields.

              GLM models do not automatically filter out attributes that may lead to the above condition. You will have to manually remove these attributes. Try just keeping the attributes with the highest coefficient values as a start and rebuilding the model.

              To determine the version of the db that you are connected to, just run the following query in sql worksheet:
              select * from v$version where banner like 'Oracle%';

              Thanks, Mark