miaodi/python-ml

Python ML Skill

Purpose

Write Python code for machine learning projects that is clear, modular, reproducible, and easy to experiment with.

When To Use

Use for model implementation, experiment scaffolding, training loops, evaluation code, and result analysis in Python-based ML projects.

Priorities

1. Preserve correctness and reproducibility.
2. Keep experiment structure simple and inspectable.
3. Make it easy to compare baselines and hyperparameters.
4. Prefer readable code over framework cleverness.

Workflow

1. Define the experiment interface: config, data flow, model, training, evaluation, and logging.
2. Keep hyperparameters centralized and easy to modify.
3. Separate reusable components from assignment-specific glue code.
4. Add clear metrics collection and result persistence.
5. Make it straightforward to run baselines, ablations, and repeated seeds.

Review Checklist

• Are configs centralized?
• Can experiments be reproduced with the same seed and settings?
• Are train/eval paths clearly separated?
• Are metrics, checkpoints, and logs easy to inspect?
• Is the code easy to adapt for ablations and sweeps?

Constraints

• Do not hide key experiment settings in scattered constants.
• Do not optimize for clever abstractions over debugging clarity.
• Keep dependencies and project structure appropriate for a course project.

Output

Provide:

• implementation plan
• code structure recommendations
• experiment hooks
• evaluation and logging guidance