PMML – The Predictive Modeling Markup Language – is the primary XML format for describing predictive analytic models so that a modeling tool can share a model with either another modeling tool or, more usefully, with a deployment environment. The folks over at KDNuggets recently ran a poll asking their readers about their use of the standard (the results are available here). While use of PMML is still pretty limited, with half the responders saying they don’t use it at all, I was encouraged to see a solid core deploying most of their models using it (15%). There were some interesting comments, too, showing the usual range from people who don’t like XML, to those who prefer deployment direct to specific code to those who like the idea of a standard but have concerns about the specific.
As someone who believes strongly that models are most valuable when both deployed and mixed with declaractive rules-based logic (for policy, regulations etc), I hope PMML will continue to gain ground. I am encouraged both by the number of modeling tools that generate it (including SAS, SPSS, Fair Isaac,IBM, R – open source) and rules-based environments that can comsume if (Fair Isaac’s Blaze Advisor, Pegasystems PegaRules and Zementis ADAPA, Chordiant Decision Management).
Any readers got comments on PMML? Anyone using it out there? I’d love to hear from you.