Imagine what happens when the creator of an operationalized models moves into a new role or leaves the organization. How do you effectively maintain, monitor, and improve upon their work?
This is where SAS Model Manager can help. This new offering helps streamline analytical model deployment and management by registering, deploying and monitoring open source models in one central environment, uniting data scientists and IT/DevOps.
SAS Model Manager allows users to ensure transparency and analytics governance. The centralized model repository, life cycle templates and version control provide visibility into your analytical processes, ensuring complete traceability and analytics governance. It enables effective collaboration by letting users track progress through each step of the model management process, and everyone involved gets a unified view of each model’s currency, definition and value.
SAS Model Manager allows you to:
- Evaluate and select the champion model from all the competing models (SAS and Open Source) for a specific business problem
- Operationalize and manage models, both SAS and open source. This means efficient model processing and governance, analytical models can be easily tested and compared, performance benchmarking reports and alerts generated, and workflow notifications sent. Modelers can collaborate and reuse models, and automatic detection notices can be sent when scoring results change over time, indicating model decay.
- Register models into a centralized repository with version control, providing visibility and simplifying compliance processes for internal governance and external regulation.
Specifically, SAS Model Manager model registration:
- Provides secure, reliable, versioned storage for all types of models, as well as access administration, including backup and restore capabilities, overwrite protection and event logging.
- Once registered, models can be searched, queried, sorted and filtered by attributes used to store them – type of asset, algorithm, input or target variables, model ID, etc – as well as user-defined propertied and editable keywords.
- Add general properties as columns to the listing for models and projects, such as model name, role, type of algorithm, date modified, modified by, repository location, description, version and keywords (tags).
- Access models and model-score artifacts using open REST APIs.
- Directly supports Python models for scoring and publishing. Convert PMML and ONNX (using SAS Python library for deep learning, called DLPy) to standard SAS model types. Manage and version R code like other types of code.
- Provides accounting and auditability, including event logging of major actions – e.g., model creation, project creation and publishing.
- Export models as .ZIP format, including all model file contents for movement across environments.
- Easily copy models from one project to another, simplifying model movement within the repository.
SAS Model Manager supports models developed with custom Python code, you can use models that are developed with packages such as scikit-learn, TensorFlow, and XGBoost. For more information about registering Python models using sasctl and PZMM module see these links below:
Looking for a way to leverage SAS Model Manager all from code? Check out a great blog entitled "Open Source Model Management through REST APIs: Registration" by Yi Jian Ching.
And see this guide by Paata U. to learn more about building and comparing open source models in SAS Model Studio.
Interested in learning about other SAS Model Manager features? Please check out other "What's new with SAS Model Manager?" posts including:
