The DataDictionary and MiningSchema elements for a PMML model requires quite a bit of metadata for each field. With sparse, high dimensional data these could each be many times larger than either the training data or the trained model.

Are there any conventional extensions (or evil non-standard kludges, depending on how you think about it) to the PMML syntax that, for instance, just says that all fields have the same metadata? Or, even better, do something like specify all fields whose names have the same prefix get the same metadata?

Also, any typical way of having the MiningSchema just say "use everything in the DataDictionary that's not the "predicted" feature as an input feature"?

  • $\begingroup$ Would all the fields need to be explicitly referenced in the models as well? If so, you'd also want to "shrink" parts of the model definition, such as the MiningSchema. $\endgroup$ – Derek Ploor Sep 3 '11 at 8:14
  • $\begingroup$ Derek - Good point. I've edited the question to make clear the same problem shows up in the MiningSchema. $\endgroup$ – DavidDLewis Sep 3 '11 at 13:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.