notebook-extraction

name	notebook-extraction
description	Use when converting Jupyter notebook experiments into reusable TeraiNet source code, removing hidden notebook state, extracting preprocessing/training/inference helpers, or adding tests for notebook logic.

Notebook Extraction

When to Use

• moving repeatable notebook logic into src/terainet/
• making notebook preprocessing, training, evaluation, or inference reusable
• removing reliance on prior cell execution, globals, or hard-coded working directories

Extraction Procedure

1. Identify the smallest self-contained behavior to extract and list its inputs, outputs, filesystem effects, configuration values, and random state.
2. Search src/terainet/ for an existing home or compatible utility before creating a module.
3. Extract pure transformations first. Use a class only when durable state is required.
4. Make paths, configuration, class mappings, preprocessing, and seeds explicit parameters.
5. Add focused tests under tests/ before replacing notebook behavior.
6. Update the notebook to import and call the extracted API; retain exploration and visualization in the notebook.
7. Compare the previous and extracted outputs on representative data, including failure cases.

Boundaries

• Preserve dataset split boundaries and label ordering.
• Do not copy notebook globals, credentials, environment-specific paths, or interactive setup into library code.
• Avoid creating broad utility modules. Place code with its closest domain: data, detection, models, or a narrowly named new module.
• Keep public APIs small, typed where practical, and documented with behavior and errors.

Completion Checklist

• No extracted function depends on prior notebook-cell state.
• Inputs, outputs, paths, configuration, and seed are explicit.
• Tests cover normal behavior and meaningful invalid input.
• Notebook behavior is preserved using the imported implementation.
• Relevant linting, type checking, and tests pass.