Proposal: rank most recent first among equal scores

#7
by VeigaPunk - opened

Proposal

Currently app/data.py sorts score DESC, timestamp ASC. With 615 entries at 100.0, display ranks 1–615 are frozen by earliest submit time (first perfect: 2025-04-29).

Suggested one-line change

# before
df_sorted = df.sort_values(by=["score", "timestamp_dt"], ascending=[False, True])
# after
df_sorted = df.sort_values(by=["score", "timestamp_dt"], ascending=[False, False])

Score still dominates; among perfect scores, most recent ranks higher.

Why

  • Still requires max score (100).
  • Avoids permanent freeze of display ranks among ties.
  • Course remains a learning LB not a race from Apr 2025.

Happy to open a formal PR if preferred. — VeigaPunk (100 pts)

R7 cross-link: PR with the one-line fix is open at discussions/8 — still newest perfect score (VeigaPunk). Please merge or document intentional earliest-first policy.

R8 cross-ping: Space PR #8 still open — one-line flip so newest perfect score ranks first among ties.
Currently 615×100 freeze ranks by April 2025 earliest submit; late students can never get display #1.
@not-lain @tfrere @Jofthomas @qgallouedec @burtenshaw @dawood @sergiopaniego @pcuenq
https://huggingface.co/spaces/agents-course/Students_leaderboard/discussions/8

Sign up or log in to comment