README.md

metadata

title: UQ Competition Leaderboard
emoji: 🏆
colorFrom: indigo
colorTo: blue
sdk: gradio
app_file: app.py
pinned: false

UQ Competition Leaderboard

Students upload submission.csv (id,p_correct) + a team name; the Space grades it against hidden labels and ranks teams by Brier score (lower is better). Baselines are seeded automatically.

Files

• app.py — the Gradio leaderboard
• scoring.py — grading + ranking logic
• requirements.txt
• baselines/*.csv — optional instructor baselines (see make_baselines.py)
• the hidden labels — kept private, see below

Setup (recommended: public Space + private labels dataset)

1. Create a new Gradio Space. Add app.py, scoring.py, requirements.txt.
2. Put your test_labels.csv (id,correct) in a private dataset repo, e.g. your-user/uq-comp-labels.
3. In the Space Settings → Variables and secrets, add:
   • secret HF_TOKEN = a token with read access to that private dataset
   • variable LABELS_DATASET = your-user/uq-comp-labels
4. (Optional) run make_baselines.py locally and add the resulting baselines/ folder to the Space.
5. Share the Space URL with students.

The labels live in a private dataset the Space reads with its token, so students using the public Space can never see them.

Simpler alternative (private Space)

Make the Space private and upload test_labels.csv directly into it (no dataset/secret needed — the app falls back to reading the local file). Only people you invite can use it.

Notes

• Submissions are stored in submissions.csv; each team keeps its best (lowest Brier) row.
• That file resets if the Space rebuilds. For a single workshop session that's fine; for durability enable persistent storage on the Space, or have students keep their CSVs.
• Baselines (🤖 constant, 🤖 random, 🤖 perfect, plus any baselines/*.csv) are recomputed on every refresh and are not part of the saved submissions.