Model Governance: Model Versioning
This article describes how ModelOp Center enables model versioning for all types of models.
Table of Contents
Introduction
ModelOp Center provides comprehensive versioning for all models under management, inclusive of versioning of all items that compose a model. For each version of a model, ModelOp Center takes a snapshot of all the elements that are part of the Standard Model Definition within ModelOp Center’s Production Model Inventory, such that you have long-term audibility and reproducibility.
Details
System of Record
All of the elements of the Standard Model Definition are versioned using ModelOp Center’s Model Manager micro service, and for external artifacts (e.g. Model Source Code), are backed by your preferred enterprise-standard versioning tooling.
For example:
Model Source Code → backed by Git (all git-compliant systems: Github, Bitbucket, Gitlab, etc.)
Model Assets → backed by standard artifact repositories (such as S3, Artifactory, etc.)
Metadata & Other Elements → backed and versioned by ModelOp Center’s Model Manager
Provenance (Traceability)
As a result, ModelOp Center provides systematic traceability of all of the core elements of a model. The immutable snapshot contains the exact asset (or reference) so that an auditor can easily trace back to the exact source that was reviewed, approved, and used for production decisioning.
Source code traceability: reference to the exact commitID that is used
Model asset traceability: reference to the exact immutable copy of the coefficients/weights/serialized-model-artifact that is used
Dependent libraries: reference to the exact list of dependent libraries that are used
Schemas: reference to the exact schema definition that is used
Job (Test): reference to the exact test jobs that were run systematically (not via a human) as part of the testing/approval process, with the ability to reproduce the tests using the snapshot information
Test Results: exact results of the systematic tests for each snapshot
Approvals: reference to each approval that was raised for a given snapshot, with traceability to the ticketing system
Model Snapshots
Behind the scenes, ModelOp Center’s data model creates a new model snapshot that is an immutable copy of all the elements (or references to the elements) that compose that model at a given point in time. These snapshots are persisted in perpetuity for long term auditability. For more details about the ModelOp Center metadata and data model, see Model Metadata Details.
To Create a New Model Snapshot
A new model version is created when a User “submits” a model for productionization. This can be done via the ModelOp Center UI or the Jupyter or RStudio plugins.
Create Version via the ModelOp Center UI
Create Version via the Jupyter Plugin
To create a new version via the Jupyter Plugin, please see the “Submit a Model” section of the Submit Model Version using Jupyter.
Create Version via the RStudio Plugin
To create a new version via the RStudio Plugin, please see the “Submit a Model” section of the Submit Model Version using RStudio.
Next Article: Model Governance: Model Metadata Details >