Zementis has recently announced its ADAPA predictive analytics edition for the Amazon Elastic Compute Cloud. This essentially allows you to deploy PMML (Predictive Model Markup Language, an XML format for defining predictive analytic models) on the Amazon compute cloud. Based on their Enterprise Edition (which has PMML deployment, reporting and business rules (using Drools), this version just allows the deployment and execution of predictive analytics models.

It uses browser based deployment, is standards based - web services for run-time access, PMML for model design and JSR 73 (Java Data Mining) for run-time access to the engine. Because it uses PMML you can use R or SPSS or SAS to build a model because these tools all support export into PMML.

It offers all the usual cloud benefits - scalable, pay as you go, reliable - as well as security through dedicated and non-shared virtual machines on the cloud - single tenant instances. It offers flexible instances - small, large, extra large - all managed from a browser-based ADAPA control center. Payment is done through the amazon.com payment engine (hence the post title) and all instances are managed through a Control Center. A simple interface allows you to pick the instance type you want and create it from the browser. You can easily create multiple instances e.g. one for QA, one for development and another for production. Once an instance is started you can load a PMML model (or several), get warnings if there are issues with the PMML, interact with the PMML to fix these issues and deploy the model. Once deployed you can use the model as a web services (for which WSDL is generated for you) or interactively. This allows you to upload and run data - either to score it or to match it against expected scores for testing. Scored data can be downloaded into Excel or anything else using CSV. Help is provided through an extensive blog with video tutorials.

This is a fascinating and quite different model for predictive analytics. Rather than having to invest in significant infrastructure to put models to work in production you are now able to use the cloud and deploy them as you go and as you like very cost effectively. Given how often the barrier to value from predictive analytics is the deployment of the model, this is a great development and I look forward to seeing what people do with it.