Model Registration in MARM

To ensure a seamless transition between development and governance, MARM eliminates the need for redundant data entry by leveraging a unified ecosystem.

To initiate any model’s validation in MARM, the model must first be available in MMG. MARM supports a pull-based approach, allowing users to onboard models from MMG for validation and monitoring.

There are two ways to register and use models:
  • Local Model Registration
  • External Model Registration
Local Model Registration
  1. If the model is developed locally:
    1. Create the model within the environment.
    2. Provide the required Pipeline Metadata, including:
      1. Dataset
      2. Model Type
      3. Target Column
      4. Prediction Column
      5. Date-Time Column
      6. Training Start Date
      7. Training End Date
    3. Publish the model pipeline.

      Once published:
      1. The model version is created and becomes a read-only model.
      2. You can pull it into MARM.
      3. The Validator can initiate the validation.
    External Model Registration

    Externally developed models can be on-boarded into MMG by creating the appropriate dataset and providing the correct training dates through the pipeline. MMG supports multiple source types, including files and tables.

    Validation data does not need to be imported separately. MMG computes validation on demand using the actual dataset, based on dataset filters and the provided training or as-of dates.

    Note:

    The current approach does not preserve historical validation results from the original external environment and assumes that the underlying data remains available for recomputation.
    If the model is coming from an external source:
    1. Register the model in MMG.
    2. Ensure all required Pipeline Metadata is provided.
    3. Additionally, register the associated datasets, as they are required for metadata selection.
    4. Publish the model pipeline.

      Once published:

      1. The model version is created and becomes a read-only model.
      2. It can be pulled into MARM for further validation and usage.
Running Example: Retail PD Model
  • Model Name: RETAIL_PD_MODEL_V1
  • Business Purpose: Predicts the likelihood of a customer defaulting on a credit card.
  • Source: Pulled directly from the MMG Modeler Workspace.
  • Validation due date: 01-April-2026