Instruction file imported from takeru110/my-github-instructions (.github/instructions/python-quality.instructions.md). Copyright stays with the author.

Python quality rules

Runtime / dependencies

• Default to the approved libraries are numpy, pandas, pyarrow, scikit-learn, torch, torchvision, Pillow, matplotlib, seaborn, hydra-core, omegaconf, optuna, hydra-optuna-sweeper, tqdm, joblib, opencv-python, scipy, statsmodels.
• If another library is clearly a better fit, you may use it.

Paths / I-O

• Use pathlib.Path for all file paths (no os.path).
• Wrap critical I/O in try/except and log stack traces via logger.exception(...).

Logging / security

• Use the logging module for status updates (no print()).
• Never log secrets, tokens, credentials, or private/internal URLs.

Code quality

• Type hints are required for all functions and methods.
• Use Google-style docstrings for public APIs and non-trivial functions/classes.
• Follow PEP 8 naming conventions and keep functions small and focused.

Data/code performance (when applicable)

• Prefer vectorized operations (pandas/numpy) over explicit Python loops.
• Use explicit indexing with loc/iloc; use groupby for aggregations when appropriate.