What is this?
Hockey Game Similarity searches NHL history to find the most similar game β or season β to what you're watching right now. It's Shazam for hockey games.
The Three Tools
Now Playing β Picks up a live game and finds the 5 most similar historical games through the same period. In the first period, team state is blended in. By period 3, the actual scoring pattern takes over.
Today's Matchups β Before puck drop, compares each team's rolling stats (GF/GA, PP%, save%, recent form) against historical matchups to find the most similar games ever played.
Season Pace β Pick any team and find which historical seasons were on the same trajectory at this point in the year. Shows how those seasons ended.
How Similarity Works
Each game is converted into a 74-dimensional vector β 74 numbers capturing everything measurable:
- Goals by period β P1, P2, P3, OT for each team (8 dims)
- Shots by period β shots on goal each period (8 dims)
- Goalie performance β save%, shots against (4 dims)
- Power play β goals, opportunities, PP% (6 dims)
- Physical play β hits, blocked shots, giveaways, takeaways (8 dims)
- Special flags β shutout, hat trick, penalty shot, comeback, OT, SO (20 dims)
- Game flow β lead changes, largest lead, comebacks, scoring pace (10 dims)
- Arena & game length β home team hash, REG/OT/SO (2 dims)
Similarity is the Euclidean distance between two vectors. 100% = identical, 0% = maximally different. In practice, 85%+ is a very strong match.
Early Period: The Blend
A 0-0 game after period 1 is somewhat informative (15% of NHL games are still 0-0 after P1), but not definitive. The live tool blends in team-state similarity in the first period:
- P1 in progress: 100% team state, 0% game state
- After P1: 50% team state, 50% game state
- After P2+: 100% game state (only 0.2% of games still 0-0)
Data Sources
- Hockey Reference β NHL box scores 1917β2009 (scoring, shots, PP, PIM)
- NHL Stats API β full game data 2010βpresent (all stats including hits, faceoffs, giveaways)