Jeremy Chen

Analytics Deployment -- Blaze Advisor on Spark

Blog Post created by Jeremy Chen Advocate on Apr 20, 2016

While the advancement in data management and analytics technologies enable organizations to explore ever increasing amount of data to extract business insight, the deployment of such business insight in a timely and transparent fashion calls for a decision management system that integrates easily with analytics assets and data sources while supporting business stakeholders’ desire to own the development and deployment of analytics. It also requires the organizations to have the demonstrable capability for tracking and monitoring the deployed analytics in production.


A powerful decision rules execution engine coupled with a highly flexible and customizable decision rules authoring capabilities, FICO Blaze Advisor is a proven leader in analytics deployment with its ability to integrate directly with multiple model formats: PMML, SAS, SRL projects produced by FICO Model Builder, and FSML (FICO Strategy Markup Language) used by FICO Analytic Modelers. With this broad set of integration support, users can import predictive and prescriptive models into Blaze Advisor with speed and fidelity.


Analytics model execution is often the last step in an organization’s pursuit to turn business insight into decisions. Before a model is executed, the organization needs to generate the variables, often in large quantity, and feed the variables to the model. Blaze Advisor supports this process from beginning to end. If new variables need to be developed, business analysts can develop them using Business Terms, including defining the calculation code and meta information used for governance purpose. Blaze Advisor also supports translating the variable definitions from SAS data steps by using one of its Add-on components called Model Translator.


Spark deployment for Blaze Advisor


While Blaze Advisor is certified on Cloudera Hadoop, Apache Spark provides another approach to run the Blaze Advisor rule engine for large scale data processing. This is especially important in the area of variable generation and analytic model execution. In a recent benchmarking test conducted by the Blaze engineering team, we (thank you Steven and Hiroshi) designed several tests to study Blaze runtime performance on Spark. The result is quite amazing: with a decision service that contains over 140 variables and 2 scorecard models, Blaze Advisor is able to process a million xml documents with a performance of around 4700tps on a 32-core system.




While deploying Blaze Advisor on Spark for batch processing may bring significant performance gain to the on-premises customers of Blaze Advisor, FICO’s cloud based Decision Management Suite v2.0 will bring Spark execution of decision services even closer to the customers. In this latest version of FICO DMS, you can develop new decision services using Decision Modeler powered by Blaze Advisor or even upload existing Blaze Advisor decision services into Decision Modeler to be executed on Spark, with great performance and very little set up work.