I got an update from Angoss recently. It has been a while since I was updated on the product – I last blogged about 7.0 which was the last major release. 7.5 is an upgrade from this version –a maintenance release – with 7.6 planned for the end of 2011 and 8.0 due out around mid 2012. Release 7.5 adds full support for Windows 7, both 32 and 64 bit, and added support for in-database analytics for Teradata (in addition to SQL Server and Netezza) as well as some other minor improvements.
In-database analytics for Teradata is an optional add-on that allows models to be built in Teradata nodes without replicating any data. Today this is read-only – models can’t be deployed back into Teradata – but the models can be generated to the various export formats supported by Angoss like SQL and PMML. Only the KnowledgeSEEKER and StrategyBUILDER features are enabled for in-database at this time – data profiling, decision trees, strategy trees, model analyzer – but Angoss is working on the additional features for StrategyBUILDER and KnowledgeSTUDIO (KnowledgeSEEKER is the base product with decision tree and other modeling capabilities while KnowledgeSTUDIO adds neural networks, scorecards and regression and StrategyBUILDER adds support for strategy trees).
Multi-core chips are supported in 7.5. Not all the product features take advantage of multi-core/multi-threading yet but the main ones like building decision trees, training regression models and auto growing trees all do. Broader support, and support for server clusters, is in the roadmap.
While not new to 7.5, StrategyBUILDER and KnowledgeSTUDIO support for predictive scorecards are worth a mention. StrategyBUILDER is an add-on for KnowledgeSEEKER and KnowledgeSTUDIO focused on building credit risk strategies – a combination of data mining, decision tree construction and business rule specification. A “strategy” in this context is a collection of business rules – typically in a decision tree format – that contains both the conditions of a segmentation and the actions to be taken on the members of those segments. StrategyBUILDER lets users develop key performance indicators for each segment in a decision tree and then assign treatment actions based on these KPIs. The resulting tree can be exported in various formats including PMML (which would let you load the tree into various business rules management systems). In fact all Angoss products (KnowledgeSEEKER, KnowledgeSTUDIO, StrategyBUILDER) automatically generate PMML for all types of models – Decision Trees, Strategy Trees, Logistic and Linear Regression, Multi-Level Neural Networks, Cluster Analysis, and Scorecards.
KnowledgeSTUDIO has supported the development of predictive scorecards since version 6.0 (November, 2007). The tool supports all the usual scorecard building steps – bringing in data, profiling and discovery (often using decision trees), coarse-grained classing, variable selection, predictive modeling, turning the model into a scorecard and finally validating and deploying it. The coarse classing algorithm generates bins –ranges of values for a variable that divide the population in meaningful ways – automatically while allowing you to consider both weight of evidence and information value and allowing manual override of bins. Standard KnowledgeSTUDIO features can be used to select the most predictive variables and build a predictive model, using logistic regression for example. A scorecard wizard is then used to turn this model into a scorecard and generate the variables, bins, weights etc. If you are not familiar with predictive scorecards, check out this great description by Eric Charpentier over on his blog.
Release 7.5 also added some new lower-level features such as scoring metrics for second nearest cluster, an ability to access the values at root and parent nodes in strategy tree nodes allows relative value calculations in tree nodes for instance, and use of SQL functions in the SQL code generator to simplify the generated SQL.