3 Replies Latest reply: Nov 22, 2012 8:30 AM by Denny Wong-Oracle RSS

    Build model - view test result - i have no ROC tab

    975581
      Dear all,

      When I build eg. a classification SVM (Lin.Reg, Naive Bayes)) model and view the test results, in that window I have only these tabs: Performance, Performance Matrix, Lift, Profit. I have no Roc tab. Do you know why? Before I used SQL Developer 3.2.09.30_x64 with this problem. Now I use SQL Developer 3.2.20.09.87_x64 with this problem. When I try to tune the SVM algoritm (so not let it automatic), there are only these tabs: Cost, Benefit, Lift, Profit. But not ROC tab :(((
      When I use the old Oracle Data Miner software from 2011, version 11.1.0.3.0 (build 11705) connecting to the same database server (using the same account), and I build a classification SVM model, I have ROC curve.

      Can anyone help me to solve this misterious problem?
      Than you!!!
        • 1. Re: Build model - view test result - i have no ROC tab
          Denny Wong-Oracle
          Hi

          The ROC result is for binary target only. How many distinct values does your target column contain?

          Thanks
          Denny
          • 2. Re: Build model - view test result - i have no ROC tab
            975581
            You are rignt, I have 6 categories/ target values. In the old Oracle Data Miner software, it was possible to set the "preferred target value" during model building, so ROC curve was generated for this target value and the "other values".
            So in SQL Developer Data Miner, is it possible to set preferred target value (in a multi categorical target value set) to get ROC curve?

            Thank you for your reply!
            • 3. Re: Build model - view test result - i have no ROC tab
              Denny Wong-Oracle
              We don't have "preferred target value" during model building in the new data miner. However, you can use the Transform node to transform your target into 2 classes (preferred target value and others). You then use the output from the Transform node as input source for your model build.

              Here is a process to transform your target into 2 classes:

              - Create a Transform node
              - In Transform node, select the target column, click "Add Transformation" icon on the toolbar
              - In the Add Transform dialog, select "Custom" Binning Type, click "Generate Default Bins" button (accept default settings)
              - In the "Custom bin values" listbox, remove the non-preferred target values (select the values and click the "Remove Transformation" icon on the toolbar)
              - Now, you should have one preferred target value in the "Custom bin values" listbox, click OK to finish

              You can now connect the Transform node to a Build node. In the Build node, select the transformed target (it should have the "_BIN" suffix in the name) as the Target for model build.

              Hope this help!

              Denny