Angoss has been selling analytic software, installed and increasingly hosted, for about 20 years. Angoss aims to satisfy advanced statisticians as well as business users with the skills and acumen to develop advanced analytics. Angoss KnowledgeSEEKER 9.0 is the latest release of their flagship product.
Release 9.0 adds automated workflow to the existing decision tree capabilities (I reviewed release 8.5 here) and allows users to build, manage and reuse model development workflows. It uses a friendly visual representation and generates documentation of the same. In addition the new release improves variable selection (characteristic analysis, measures of predictive power), has a new Weight of Evidence optimizer and can generate SAS code for a node or the full workflow.
Workflow templates are supported, allowing teams to develop standard approaches. A complete workflow can also be applied to a new dataset, creating a new decision tree from a new dataset using a standard workflow. The workflow tools run in their Windows install in either client-server or single user installs.
Workflows can handle various functions:
- Datasets can be brought in from a wide range of sources – everything from text to SAS, SPSS or R data files to an ODBC database. These datasets can be explored with the usual kind of reporting, variable distribution graphs, statistical assessments etc. Specific segments based on an attribute in the dataset can be compared to each other and the dataset as a whole to find the most predictive variables. The tool supports automated binning (finding the best value ranges for differentiating) and provides tools for merging and analyzing bins as well as for determining that variables are highly correlated.
- Data transformation steps can create expressions, partitions, joins, aggregates etc. These tools have some suggestions and recommendations built in but are primarily manual, allowing an expert user to do what they need to do.
- Modeling technique nodes include Decision Trees and Strategy Trees (the classic Angoss trees). The Angoss tree techniques are very interactive and at each step there is an option to automatically find the next best split. Attributes can be excluded, maximum number of bins specified and more as part of building the tree.
- Model validation nodes can score datasets to feed into analysis nodes that check the model is valid and predictive. Multiple validation datasets can be fed into a single analysis node to compare, for instance, development and validation datasets. Lift charts, decile charts, profit curves etc can all be generated.
- Deployment nodes include scoring, generating SQL, SAS, structured English, PMML etc.
- Finally data can be exported to various data formats.
SAS code can be generated for all or part of any workflow so that the workflow can be deployed as SAS code. Multiple workflows can be stored in a project with the ability to connect projects to each other coming soon. Each project is a folder that contains a whole series of XML files and other files can be managed in the project also.
Usability remains important in KnowledgeSEEKER, especially in the tree builder, with the ability to split the screen, showing different parts of the tree in different windows, as well as overview maps, charts showing variation between nodes in the tree as the user navigates.
Angoss is one of the vendors in our Decision Management Systems Platform Technologies Report and more information on KnowledgeSEEKER is available here.