Prediction Competition Leaderboard
Higher scores are better. 1.0 indicates a theoretical perfect score (no model is likely to get
anywhere near this) and 0.0 means a model is doing as well as the reference model I'm using, which
for any game state predicts the Spy winrate for that venue in SCL6.
Sharing code/ideas is highly encouraged! Similarly, submitting models that tried new ideas that
didn't work out so well is encouraged, so that you can share your experience.
| # |
Submitter |
Score |
Model type |
Description |
| 1 |
OpiWrites |
0.113990 |
Gradient Boosting |
Same as previous model, but microfilm animation type included, as well as a value for approximate sniper time and a flag for when the clock is hanging. |
| 2 |
OpiWrites |
0.105836 |
Gradient Boosting |
GradientBoostingClassifier library from SciKit Learn, using data extracted from timelines as various feature inputs. Features include all forms of mission progress, sniper lights, time remaining, and tracking of various spy actions. Additionally, faux-memory values that give the model time elapsed since most spy actions. |
| 3 |
Tonyl |
0.050341 |
Expert System |
Predicts based on the average win rate for the combination of venue, spy light and number of hard tells completed. |
| 4 |
Wobble |
0.039970 |
Expert system |
Predicts 0.3, 0.45 or 0.8 depending on spy HL status |
| 5 |
Wobble |
0.000000 |
Expert system |
Uses the SCL6 spy win rate for its prediction |
| 6 |
Wobble |
-0.007664 |
Expert system |
Always predicts 0.45 |