SAS^® Model Manager

Features

Central, secure repository for managing analytical models

• Project-based storage of models.
  • Set up and maintain separate versions of champion and challenger models within a project:
    • Champion model is automatically set as a default version. One only champion model is produced per project.
    • Select challenger models to the project champion model.
    • Monitor and publish challenger and champion model packages.
  • Map prerequisite data sources used for model reporting and score code testing.
  • Accounting and auditability:
    • Event logging of all major actions.
    • User-defined notes.
    • Ability to attach documents (Microsoft Word documents, Microsoft Excel spreadsheets, HTML files, etc.), and add version control.
• Prebuilt templates for registering standard data mining models, including prediction, segmentation, classification and scorecards.
• User-defined templates.
• Optional batch model registration support for bulk loading.
• General properties such as model name, type of algorithm, creation date, modification date, etc.
• Advanced view of SAS^® Enterprise Miner™ process flow diagram.
• Model validation reports are provided for Basel II risk models, including probability of default (PD) and loss given default (LGD).
• Provide more control in setting input and output variables to define the project.
• Import multiple Base SAS, SAS/STAT^® and SAS Enterprise Miner models, including training code, score logic, estimate tables, and target and input variables.
• Import select SAS/STAT linear models from a SAS package file (.SPK), including LOGISTIC, GENMOD, REG, GLMSELECT, GLIMMIX and MIXED.
• Import select SAS High-Performance Analytics Server models from a SAS package file (.SPK), including HPBIN, HPREDUCE, HPNEURAL, HPLOGISTIC and HPREG.
• Import and export PMML model code with inputs and outputs. Create DATA step score code for PMML models for inclusion in scoring tasks, reporting and performance monitoring.
• Register, compare, report, score and monitor models built in R.
• Repository metadata summary report with information such as number of models, number of scoring jobs, model-aging profiles, and frequency counts of how often each target and input variable has been used across the model portfolio.
• Model repository can be queried by attributes used to store models such as type of algorithm, input or target variables, model creator, model ID, etc.
• Secure, reliable model storage and access administration, including backup and restore capabilities, overwrite protection, event logging, and user authentication/access privilege administration.

Analytical workflow management

• Create custom processes for each model using SAS Workflow Studio – a Web-based client:
  • SAS Workflow Studio is used to design the model approval process that is imported and managed through the SAS Model Manager Workflow Console.
  • Provide collaboration across teams with automated notifications.
  • Define, manage and track complete analytic life cycles.
  • Enable enterprise access and collaboration with the Web interface.
  • Increase efficiency with process management capabilities.
  • Associate milestones with activities as part of the workflow process definition.
• Perform common model management tasks using the SAS Model Manager Workflow Console:
  • Import, view and attach supporting documentation and publish models.
  • Set a project champion model and flag challenger models.
  • Publish models for scoring purposes.
  • View dashboard reports.

Scoring-logic validation before models are exported to production

• Define test and production score jobs using required inputs and outputs:
  • Map required inputs and outputs.
  • Add SAS code.
  • View log and results tables.
  • Create interactive graphs.
• Scoring task scheduler:
  • Schedule scoring tasks to run at certain times and dates on available servers.
  • Specify where to save the scoring task output and view job history.
  • Export models to SAS Metadata Repository.
• Production scoring:
  • Model Scoring Task is available in SAS Data Integration Studio and SAS^® Enterprise Guide^®.
  • Publish models directly to SAS Real-Time Decision Manager.
• Publish model updates to different scoring channels:
  • Notify subscribers via email.
  • Store results to a file system or post to a corporate intranet.
• In-database model deployment:
  • Using integration with SAS Scoring Accelerator, publish and validate scoring functions for native scoring within databases.
  • Publish model scoring files using a vendor-defined function in DB2, EMC Greenplum, IBM Netezza and Teradata.
  • Publish model scoring files using the SAS Embedded Process in Aster, DB2, Oracle and Teradata.

Monitoring and reporting on model performance during test and production life cycles

• Model performance reports:
  • Variable distribution plots, characteristic charts, stability charts, lift charts, Receiver Operating Curve (ROC) charts, Kolmogorov-Smirnov (K-S) charts and Gini charts.
  • For prediction model function that has an interval target.
  • For champion and challenger model comparisons.
• Model comparison reports:
  • Model profile report, delta report, dynamic lift report, interval target variable report, etc.
  • Ad hoc SAS code report editor.
  • HTML, RTF, PDF and Microsoft Excel output formats.
  • Aggregated report to combine multiple reports from the Reports folds into a single report.
• Training summary data set report showing frequency and distribution charts to validate data set variables.
• Easy-to-use wizard for creating performance-monitoring dashboards:
  • Update all reports or update reports for projects that have new performance data.
• Model retraining allows users to create new challenger models based on SAS Enterprise Miner models currently registered in a project, and new data and variables.
• Perform scoring and performance monitoring on a database appliance (Teradata or EMC Greenplum) that has been configured for use with SAS High-Performance Analytics Server:
  • Calculate model performance statistics for classification and prediction models.

Overall lifecycle management of analytical models

• Model lifecycle templates for collaborative project management:
  • Basic, standard, extended and user-defined.
  • Model Lifecycle Template editor for user-defined templates.
  • Task-oriented milestone completion and approval signoff.
• Progress-completion status reports.