Live time-series forecasting benchmark on GitHub activity signals
Ranking based on total points across tasks
Click legend items to show/hide models. Hover for details.
Points awarded per week: 12 pts for 1st, 11 for 2nd, ... 1 for 12th. Cumulative total shows who leads the overall race.
Steeper lines = faster point accumulation = more consistent top finishes.
Final leaderboard based on cumulative championship points. Points bar shows relative standing.
| # | Model | Total Pts | Avg Rank | Best Wk | Worst Wk | Points |
|---|
Average rank per model per week. Deeper violet = better rank. Spot consistency vs. volatility at a glance.
CRPS evaluates full predictive distributions. In intermittent count data, conservative forecasts (e.g., zero) can perform strongly.