IModelRegistry<T, TInput, TOutput>

Interfaces · AiDotNet.Interfaces

Defines the contract for model registry systems that manage trained model storage and versioning.

Known implementations: ModelRegistryBase<T, TInput, TOutput>

How It Works

A model registry serves as a centralized repository for trained models, managing their lifecycle from development through production deployment.

For Beginners: Think of a model registry like a library for your trained models. Just like a library catalogs books and tracks which are checked out, a model registry:

  • Stores all your trained models in one place
  • Tracks different versions of the same model
  • Records which models are being used in production
  • Helps you find and retrieve the right model when you need it

Key features include:

  • Model versioning (keeping track of model evolution)
  • Metadata tracking (when trained, by whom, with what data)
  • Stage management (development, staging, production)
  • Model comparison and lineage tracking

Why model registries matter:

  • Prevents losing track of trained models
  • Enables rollback to previous versions if needed
  • Provides audit trail for compliance
  • Facilitates collaboration between team members
  • Simplifies deployment process

Methods

Method Summary
List<ModelVersionInfo<T>> ListModelVersions(string modelName) Lists all versions of a specific model.
List<RegisteredModel<T, TInput, TOutput>> SearchModels(ModelSearchCriteria<T> searchCriteria) Searches for models based on criteria.
List<string> ListModels(string filter, Dictionary<string, string> tags) Lists all models in the registry.
ModelCard GenerateModelCard(string modelName, int version, string developers) Generates a Model Card from the registered model’s metadata and evaluation results.
ModelCard GetModelCard(string modelName, int version) Gets the Model Card for a registered model version.
ModelComparison<T> CompareModels(string modelName, int version1, int version2) Compares two model versions.
ModelLineage GetModelLineage(string modelName, int version) Gets the lineage information for a model (how it was created).
RegisteredModel<T, TInput, TOutput> GetLatestModel(string modelName) Gets the latest version of a model.
RegisteredModel<T, TInput, TOutput> GetModel(string modelName, int? version) Retrieves a specific model version from the registry.
RegisteredModel<T, TInput, TOutput> GetModelByStage(string modelName, ModelStage stage) Gets the model currently in a specific stage (e.g., production).
int CreateModelVersion(string modelName, IModel<TInput, TOutput, TMetadata> model, ModelMetadata<T> metadata, string description) Creates a new version of an existing model.
string GetModelStoragePath(string modelName, int version) Gets the storage location for a model version.
string RegisterModel(string name, IModel<TInput, TOutput, TMetadata> model, ModelMetadata<T> metadata, Dictionary<string, string> tags) Registers a new model in the registry.
void ArchiveModel(string modelName, int version) Archives a model version.
void AttachModelCard(string modelName, int version, ModelCard modelCard) Attaches a Model Card to a registered model version.
void DeleteModel(string modelName) Deletes all versions of a model.
void DeleteModelVersion(string modelName, int version) Deletes a specific model version.
void SaveModelCard(string modelName, int version, string filePath) Saves the Model Card for a model version to a file.
void TransitionModelStage(string modelName, int version, ModelStage targetStage, bool archivePrevious) Transitions a model version to a different stage.
void UpdateModelMetadata(string modelName, int version, ModelMetadata<T> metadata) Updates the metadata for a model version.
void UpdateModelTags(string modelName, int version, Dictionary<string, string> tags) Adds or updates tags for a model.