Signers_stories

US Model Signature Generation : Generate signatures for AI/ML models.

• US Model Signature Generation : Generate signatures for AI/ML models.
  • classes relations
  • User Stories: Signer Management
    • 1. User Story: Configure Model Signer
    • 2. User Story: Generate Model Signatures
    • 3. User Story: Implement Signature Inference
    • Common Acceptance Criteria
    • Definition of Done (DoD):
  • Code location
  • Test location

---

classes relations

classDiagram
    direction LR

    class Signer {
        <<abstract>>
        +KIND: str
        +sign(inputs: schemas.Inputs, outputs: schemas.Outputs) Signature*
    }
    Signer --|> pdt.BaseModel : inherits
    Signer --|> abc.ABC : inherits

    class InferSigner {
        +KIND: T.Literal["InferSigner"] = "InferSigner"
        +sign(inputs: schemas.Inputs, outputs: schemas.Outputs) Signature
    }
    InferSigner --|> Signer : inherits

    class Signature {
        <<type>>
        ms.ModelSignature
    }

    class schemas.Inputs {
        <<external>>
    }

    class schemas.Outputs {
        <<external>>
    }

    Signer --> schemas.Inputs : "uses"
    Signer --> schemas.Outputs : "uses"
    Signer --> Signature : "returns"
    InferSigner --> mlflow.models.infer_signature : "uses"

Loading

User Stories: Signer Management

---

1. User Story: Configure Model Signer

Title:
As a data scientist, I want to configure a model signer to generate the model signatures, so that I can validate input and output formats after training.

Description:
The Signer abstract class allows for creating different signing strategies for models, enabling flexibility in how signatures are generated.

Acceptance Criteria:

• The signer class can be configured with specific signing strategies.
• Default implementations may be available while allowing users to define custom strategies.

---

2. User Story: Generate Model Signatures

Title:
As a data engineer, I want to generate model signatures based on the input and output structures, so that I can ensure the integrity of model interfaces.

Description:
The sign method of the Signer class generates a model signature using the provided inputs and outputs.

Acceptance Criteria:

• The method accurately captures the information from inputs and outputs to create the signature.
• The generated signature is usable in downstream applications, such as model deployment or tracking.

---

3. User Story: Implement Signature Inference

Title:
As a data scientist, I want to use the built-in inference capabilities to automatically generate signatures from model inputs and outputs, so I can simplify the signature generation process.

Description:
The InferSigner class provides functionality that leverages the MLflow library to automatically infer signatures based on the model's data.

Acceptance Criteria:

• Model signatures are generated accurately using the InferSigner strategy.
• The inference correctly captures the structure of both inputs and outputs based on the received data.

---

Common Acceptance Criteria

1. Implementation Requirements:

   • The Signer and its child classes are structured in a way that supports flexibility and extensibility of signing strategies.
   • Clear abstractions are provided for signing methods to ensure consistent interface usage.

2. Error Handling:

   • Any issues encountered during the signature generation process are logged, with informative messages describing the nature of the error.

3. Testing:

   • Unit tests validate the creation and functionality of different signer implementations.
   • Tests ensure that signatures reflect accurately based on provided inputs and outputs.

4. Documentation:

   • Each method and class should contain comprehensive docstrings to describe how they function.
   • Examples demonstrating how to use both Signer and InferSigner should be included for user guidance.

---

Definition of Done (DoD):

• The Signer and InferSigner classes are fully implemented and functioning as intended.
• All user stories are tested, and acceptance criteria are met.
• Documentation is comprehensive and provides clear instructions on how to utilize signing functionality.

Code location

src/regression_model_template/utils/signers.py

Test location

tests/utils/test_signers.py