Dota 2 Draft Odds
Pick up to five heroes for each team. The page estimates Radiant's win chance from how those heroes do overall and against each other in public matches, using a model that is published only after it beats a no-draft guess on matches it was not fitted on.
Details
Source: OpenDota public matches. These statistics are not filtered by patch; the patch shown is the latest one OpenDota lists. How we calculate thisPick the teams
Pick at least one hero for each team to see the odds.
ADVICE
Use it to argue about drafts, not to bet: a 55% draft still loses almost half the time.
Was it right? Check a finished match
Enter the ID of a finished public match (from the match page in the game, OpenDota or Dotabuff). The page shows what the model gave that draft and who won; checked matches add to the public tally below.
DATA
No matches have been checked with this model yet.
Calibration on held-out matches
| Metric | Model | Side-only guess |
|---|---|---|
| Log loss (lower is better) | 0.6740 | 0.6865 |
| Brier score (lower is better) | 0.2406 | — |
Fitted on 34,508 older matches, tested on the 8,627 newest. Both scores measure how far the stated odds were from what actually happened, so lower is better. The model is shown only while it scores better than the side-only guess.
| Predicted | Actual | Matches |
|---|---|---|
| 27.7% | 0.0% | 4 |
| 37.1% | 37.3% | 166 |
| 46.3% | 45.2% | 1,691 |
| 55.2% | 55.4% | 3,887 |
| 63.9% | 62.7% | 2,539 |
| 72.5% | 71.3% | 338 |
| 80.9% | 50.0% | 2 |
How the odds are calculated
INTERPRETATION
- Each hero's overall win rate and every Radiant-versus-Dire hero pair's result in public matches (OpenDota, not filtered by patch).
- A pair counts only by how far it differs from what the two heroes' overall win rates predict; pairs with few games are pulled toward zero.
- Three weights (overall strength, pair effects, side) are fitted on older public matches and tested on the newest ones.
- Not included: allies' synergy (no data source), player skill, rank, roles and lanes. That is why the number is an estimate, not a prediction.