Skip to content

Requirements⚓︎

You can find here a detailed list of requirements, clearly specifying the desired features and behaviors that the system must meet.

MLOps requirements
MLOps requirements satisfaction

Info

This analysis is a work in progress, and will evolve. 👷

Model Trainer⚓︎

  • BR086: The Model Trainer shall support management of machine learning training - covering models developed in the most popular machine learning frameworks, such as TensorFlow, PyTorch, Keras, etc.
  • BR087: The Model Trainer shall support management of machine learning models using interoperable model representations, such as ONNX.
  • BR088: The Model Trainer shall support initiation of training runs and provide the resultant metrics to assess model performance.
  • BR089: The Model Trainer shall maintain a history of runs, including associated parameterization, training datasets and model metrics. Each run should be associated with the persisted model output assets.
  • BR090: The Model Trainer shall persist model assets within platform storage, including integration within user/project workspace.
  • BR091: The Model Trainer shall support applying versions to generated models.
  • BR092: The Model Trainer shall support registration of versioned models within Resource Discovery - e.g. to publish for sharing.
  • BR093: The Model Trainer shall provide a Web UI with access to the training run history, through which model runs can be managed and assessed for performance [BR097-1] .
  • BR094: The Model Trainer shall support integration as service within the Workspace BB.

Training Data Manager⚓︎

  • BR095: The Training Data Manager shall support version-controlled management of training datasets. The approach to training dataset management should facilitate discovery and reuse, by use of appropriate metadata standards such as STAC extensions [Resource Discovery RD23].
  • BR096: The Training Data Manager shall support registration of versioned ML training data within Resource Discovery - e.g. to publish for sharing.

MLOps UI⚓︎

  • BR097: The MLOps UI shall provide a web-enabled UI that provides:
    • BR097-1: Access to the training run history, through which model runs can be managed and assessed for performance.
    • BR097-2: Access to training data, for purposes of management, versioning etc.