Today I'm trying to learn how the ThingPredictor works. I try the demo (Demo: Using ThingPredictor) with the JetEnginePredictiveModel and everything seems fine (I do the tut up to step 5).
Now I try it with my own models (and the vibration dataset from the Analytics Builder Quickstart). In Analytics Builder I select the model and click on the publish button. After some seconds a new tab opens with the Analytics Manager and the opened Analytics Model. I do the step 4 of the Using ThingPredictor this model (including selected parant row (step 4.c.d.) with message "Successfully set Parent"). Add some values (step 4.c.f.) and the row and submit the job. Than I wait are short time. Click on refresh, select the "predictiveScore" in the Result Data Shape treeview and get this error message:
The same message appears when I click on importantFields. It look likes I haven't done step 4.c.d. but I do and I get the message "Successfully set Parent".
Ok I get some clue about the error.
On the Analytics Jobs list I see the job state is FAILED. If I click view in the Analytics Job list I saw the message "Cannot compute important fields on a non-numeric goal". Now I try is with an numeric field and it work but the field I will monitor is an booleanfield. How can I do this with the ThingPredictor?
Are you setting an Important Field when attempting to run a Boolean goal?
If so, this field is optional, and may cause issues with running ThingPredictor scoring on Boolean Goals. Please try to run it again without Important Fields.
Also, I would like to provide our Online HelpCenter around ThingPredictor, which is updated on an on-going basis; it may contain some useful information for you if you have any other questions.
I can indeed reproduce this error when using a Boolean goal.
I have reproted this to R&D, see Article Viewer | PTC for reference.
It does work though when using a numeric goal, but for the moment I do not see a workaround when using Boolean goal.
Ok, I hope this will fixed soon.
One workaround at the moment I use is to convert all boolean goals to number with value 0 or 1. But I can not test the results of this because of the ROC and Confustion Matrix failed with wrong values problem.