Inference_stories

US Model Inference Job : Define a job for generating batch predictions from a registered model.

• US Model Inference Job : Define a job for generating batch predictions from a registered model.
  • classes relations
  • User Stories: Inference Job Management
    • 1. User Story: Configure Inference Job
    • 2. User Story: Read Input Data
    • 3. User Story: Load the Model
    • 4. User Story: Generate Predictions
    • 5. User Story: Write Predictions to Data Output
    • 6. User Story: Notify Completion of Inference
    • Common Acceptance Criteria
    • Definition of Done (DoD):
  • Code location
  • Test location

---

classes relations

classDiagram
    class InferenceJob {
        +KIND: T.Literal["InferenceJob"] = "InferenceJob"
        +inputs: datasets.ReaderKind
        +outputs: datasets.WriterKind
        +alias_or_version: str | int = "Champion"
        +loader: registries.LoaderKind = CustomLoader()
        +run() base.Locals
    }

    class Job {
        <<abstract>>
        +run()* base.Locals
    }
    InferenceJob --|> Job : inherits

    class ReaderKind {
        <<type>>
        Reader
    }

    class WriterKind {
        <<type>>
        Writer
    }

    class LoaderKind {
        <<type>>
        Loader
    }

    class CustomLoader {
        +load(uri: str) Any
    }
    CustomLoader --|> LoaderKind : implements

    InferenceJob --> ReaderKind : "uses"
    InferenceJob --> WriterKind : "uses"
    InferenceJob --> LoaderKind : "uses"

Loading

User Stories: Inference Job Management

---

1. User Story: Configure Inference Job

Title:
As a data scientist, I want to configure an inference job that specifies the necessary parameters for generating predictions, so that batch predictions can be effectively processed.

Description:
The InferenceJob class enables the setup of the job with parameters such as input data readers, output data writers, model details, and the loader for accessing the model.

Acceptance Criteria:

• The job is initialized with the necessary parameters.
• Default values are properly handled for optional fields.

---

2. User Story: Read Input Data

Title:
As a data engineer, I want to read input data from specified sources, so that the model can generate predictions based on these inputs.

Description:
In the run method, the input data is read using the designated data reader, which ensures data integrity and prepares it for prediction.

Acceptance Criteria:

• The job successfully reads input data using the configured reader.
• Input data is validated and conforms to the expected schema.

---

3. User Story: Load the Model

Title:
As a data scientist, I want to load the registered model from the model registry, so that I can use it to generate predictions on the input data.

Description:
The job uses the configured loader to access the specified version or alias of the model from the registry.

Acceptance Criteria:

• The model is loaded correctly from the registry using the provided loader.
• The model instance must be ready for making predictions.

---

4. User Story: Generate Predictions

Title:
As a data scientist, I want to generate predictions using the loaded model and the input data, so that I can evaluate the model's performance on new data points.

Description:
The job leverages the model's predict method to produce output predictions based on the input data.

Acceptance Criteria:

• Predictions are generated using the loaded model and validated input data.
• The output of the predictions is in a usable format for further processing.

---

5. User Story: Write Predictions to Data Output

Title:
As a data engineer, I want to write the generated predictions to a specified data output, so that results can be stored and retrieved later.

Description:
The job takes the prediction outputs and writes them to the designated storage using the configured writer.

Acceptance Criteria:

• The predictions are successfully written to the specified output using the writer.
• The method of storage should ensure data integrity.

---

6. User Story: Notify Completion of Inference

Title:
As a user, I want to be notified when the inference job is finished, along with the shape of the output data, so that I can review the results promptly.

Description:
At the end of the job execution, notifications are sent to relevant stakeholders summarizing the outcome, including predictions shape.

Acceptance Criteria:

• Notifications include job completion details, specifically the shape of the outputs.
• The alerts service successfully informs users about job completion status.

---

Common Acceptance Criteria

1. Implementation Requirements:

   • The InferenceJob class correctly implements the abstract run method from the base Job class.
   • All necessary services (logging, model registry, alerts) are initialized at the start of the inference job.

2. Error Handling:

   • Clear error messages are logged for any issues encountered during the reading, loading, or writing processes.

3. Testing:

   • Unit tests validate job initialization, data reading, model loading, prediction generation, and output writing.
   • Tests ensure that errors in processes trigger appropriate logging and notifications.

4. Documentation:

   • Each class and method in the InferenceJob should have clear docstrings and examples provided for clarity.
   • Users should be guided on how to configure and use the inference job.

---

Definition of Done (DoD):

• The InferenceJob class is fully implemented and tests pass all acceptance criteria.
• The functionality includes reading inputs, loading models, generating predictions, writing outputs, and notifying users.
• The documentation is complete and well-structured for ease of understanding.

Code location

src/regression_model_template/jobs/inference.py

Test location

tests/jobs/test_inference.py