Homework 1

Prediction Competition

Author

Term 2 - HW 1

Goal

Participate in the course prediction competition and build a reproducible prediction system rather than optimizing a single leaderboard submission.

Requirements

  • Define the outcome, unit of prediction, loss/leaderboard metric, and the information available at prediction time.
  • Create a leakage-resistant train/validation strategy before tuning models. Explain why it matches the data-generating or deployment setting.
  • Submit at least two substantively different baselines and one improved model. Report validation performance, calibration or residual diagnostics as appropriate, and a brief error/slice analysis.
  • Save code, package/environment information, predictions, and a README.md with exact reproduction instructions. Do not include hidden test labels in the repository.
  • Write a 2-page model card: intended use, inputs, validation design, results, limitations, and potential harms or failure modes.

AI agents may help with API questions or debugging. Record prompts and output relied on in AI_USE.md, and independently verify that no suggested feature leaks future or target information.