Deep learning system for data-driven skill estimation
Various aspects of the subject technology relate to systems, methods, and machine-readable media for determining player skill for video games. The method includes aggregating a plurality of player statistics for match outcomes from a plurality of video games. The method also includes calculating, for each player in a pool of players, a matchmaking rating for each player based on the plurality of player statistics, the matchmaking rating for each player comprising a predicted number of points each player will contribute to a match. The method also includes selecting, based on the matchmaking rating for each player, players from the pool of players. The method also includes matching the players based on the matchmaking rating for each player, a sum of the matchmaking ratings comprising a total predicted team score for the match.
1. A computer-implemented method for determining player skill for video games, comprising:
aggregating a plurality of player statistics from a plurality of video games;
determining, for each player in a pool of players, a skill rating for each player based on a machine learning algorithm comprising a neural network that uses the plurality of player statistics, wherein the neural network is a Siamese neural network that comprises a branch for each player; and
matching the players in the pool to a team based on the skill rating for each player.
2. The computer-implemented method of claim 1 , wherein determining a skill rating for each player comprises training the machine learning algorithm based on the plurality of player statistics to determine a correlation between the plurality of player statistics and a match outcome.
3. The computer-implemented method of claim 1 , further comprising updating the machine learning algorithm based on a comparison between the predicted number of points for a team point total and an actual number of points for a team point total of a match outcome.
4. The computer-implemented method of claim 1 , further comprising matching a first team to a second team to participate in a contest or match based on a match making rating for each player on a first team and a second team.
5. The computer-implemented method of claim 4 , wherein a sum of the match making ratings of each player of the first team is equal to a sum of the match making ratings of each player of the second team.
6. The computer-implemented method of claim 1 , wherein the skill rating for each player is based on a number points each player on the team contributes to a team total.
7. The computer-implemented method of claim 1 , wherein the skill rating for a team comprises a predicted number of points each team will score in a contest or match.
8. The computer-implemented method of claim 1 , wherein the pool of players request to participate in a contest or match.
9. The computer-implemented method of claim 1 , wherein the plurality of video games comprises different video games.
10. The computer-implemented method of claim 1 , wherein the training is based at least in part on contextual data comprising at least one of in-game maps, game modes, and/or player roles.
11. A system for determining player skill for video games, comprising: a processor; and
a memory comprising instructions stored thereon, which when executed by the processor, causes the processor to perform:
aggregating a plurality of player statistics from a plurality of video games;
determining, for each player in a pool of players, a skill rating for each player based on a machine learning algorithm comprising a neural network that uses the plurality of player statistics, wherein the neural network is a Siamese neural network that comprises a branch for each player;
matching the players in the pool to a team based on the skill rating for each player; and
updating the machine learning algorithm based on a comparison between a predicted number of points for a team point total and an actual number of points for a team point total of a match outcome.
12. The system of claim 11 , wherein determining a skill rating for each player comprises training the machine learning algorithm based on the plurality of player statistics to determine a correlation between the plurality of player statistics and a match outcome.
13. The system of claim 11 , wherein the processor further performs matching a first team to a second team to participate in a contest or match based on a match making rating for each player on a first team and a second team.
14. The system of claim 13 , wherein a sum of the match making ratings of each player of the first team is equal to a sum of the match making ratings of each player of the second team.
15. The system of claim 11 , wherein the skill rating for each player is based on the number points each player on a team contributes to a team total.
16. The system of claim 11 , wherein the skill rating for a team comprises a predicted number of points each team will score in a contest or match.
17. The system of claim 11 , wherein the plurality of video games comprises different video games.
18. The system of claim 11 , wherein the training is based at least in part on contextual data comprising at least one of in-game maps, game modes, and/or player roles.
19. A non-transitory computer-readable storage medium comprising instructions stored thereon, which when executed by one or more processors, cause the one or more processors to perform operations for determining player skill for video games, the operations comprising:
aggregating a plurality of player statistics from a plurality of video games;
determining, for each player in a pool of players, a match making rating for each player based on a machine learning algorithm comprising a neural network that uses the plurality of player statistics wherein the neural network is a Siamese neural network that comprises a branch for each player;
matching the players in the pool to at least a first team and to a second team to participate in a contest or match based on a match making rating for each player on a first team and a second team; and
updating the machine learning algorithm based on a comparison between a predicted number of points for a team point total and an actual number of points for the team point total of a match outcome.