SYSTEMS AND METHODS FOR A TRANSFORMER NEURAL NETWORK FOR PLAYER AND TEAM PREDICTIONS FOR SPORTS
A method for generating predictions for teams and players associated with a sporting event using a transformer neural network, the method including: receiving a set of input features, the set of input features representing a set of players and teams within a match, each input within the set of input features being represented by a tensor; inputting the set of input features into a transformer neural network, the transformer neural network including: a set of embedding layers; transformer encoder layers; and fully connected layers; and generating, using the transformer neural network, a set of target metric predictions for the set of players and teams within the match.
1 . A method of generating predictions for teams and players associated with a sporting event using a transformer neural network, the method comprising:
receiving a set of input features, the set of input features representing a set of players and teams within a match, each input within the set of input features being represented by a tensor;
inputting the set of input features into a transformer neural network, the transformer neural network including:
a set of embedding layers;
transformer encoder layers; and
fully connected layers; and
generating, using the transformer neural network, a set of target metric predictions for the set of players and teams within the match.
2 . The method of claim 1 , wherein each tensor within the set of input features corresponds to a grain level of a plurality of a grain levels, the grain level indicating that the tensor belongs to either a player-level, a team-frame-level, or a game-level category.
3 . The method of claim 1 , wherein the set of input features includes:
a first tensor of live player features indicating a player's position and team, and a running total of actions performed by a player in the set of players;
a second tensor of player strength features indicating a player's aggregate statistics of the player's actions over a set of previous games;
a third tensor of live team features indicating running totals of actions performed by a team of the set of teams within the match;
a fourth tensor of team strength features indicating aggregated statistics of a team's aggregate statistics over the set of previous games;
a fifth tensor of live game state features indicating attributes of the match including at least one of an event-type, a game-clock time, or event location; and
a sixth tensor of game context features indicating a league in which the match is taking place and a time associated with the match.
4 . The method of claim 1 , wherein inputting the set of input features into the transformer neural network further includes:
mapping, using the set of embedding layers, the set of input features into a set of tensors with a common feature dimension, wherein the set of embedding layers includes a linear layer for each input of the set of input features.
5 . The method of claim 4 , wherein inputting the set of input features into the transformer neural network further includes:
the transformer encoder layers receiving the mapped set of tensors with common feature dimension from the set of embedding layers; and
computing, using the transformer encoder layers, self-attention along temporal and agent dimensions to the mapped set of tensors with common feature dimensions to generate transformer encoder layer embeddings.
6 . The method of claim 5 , wherein inputting the set of input features into the transformer neural network further includes:
mapping, using the fully connected layers, the transformer encoder layer embeddings into tensors corresponding to target metrics.
7 . The method of claim 1 , wherein creating a set of inputs features occurs automatically upon detection of the sporting event being scheduled.
8 . The method of claim 7 , wherein a feature creator processing step initiates creation of the set of input features for the sporting event, upon querying data from a data platform.
9 . The method of claim 1 , wherein the target metric predictions include: goals, assists, shots, shots on target, passes, fouls, yellow cards, red cards and/or minutes for one or more of the players.
10 . The method of claim 1 , wherein the set of generated predictions for at least one action specific for each player or team associated with the sporting event is updated temporally during a match.
11 . A system for generating predictions for teams and players associated with a sporting event using a transformer neural network, the system comprising:
a non-transitory computer readable medium configured to store processor-readable instructions; and
a processor operatively connected to the non-transitory computer readable medium, and configured to execute the instructions to perform operations comprising:
receiving a set of input features, the set of input features representing a set of players and teams within a match, each input within the set of input features being represented by a tensor;
inputting the set of input features into a transformer neural network, the transformer neural network including:
a set of embedding layers;
transformer encoder layers; and
fully connected layers; and
generating, using the transformer neural network, a set of target metric predictions for the set of players and teams within the match.
12 . The system of claim 11 , wherein each tensor within the set of input features corresponds to a grain level of a plurality of a grain levels, the grain level indicating that the tensor belongs to either a player-level, a team-frame-level, or a game-level category.
13 . The system of claim 11 , wherein the set of input features includes:
a first tensor of live player features indicating a player's position and team, and a running total of actions performed by a player in the set of players;
a second tensor of player strength features indicating a player's aggregate statistics of the player's actions over a set of previous games;
a third tensor of live team features indicating running totals of actions performed by a team of the set of teams within the match;
a fourth tensor of team strength features indicating aggregated statistics of a team's aggregate statistics over the set of previous games;
a fifth tensor of live game state features indicating attributes of the match including at least one of an event-type, a game-clock time, or event location; and
a sixth tensor of game context features indicating a league in which the match is taking place and a time associated with the match.
14 . The system of claim 11 , wherein inputting the set of input features into the transformer neural network further includes:
mapping, using the set of embedding layers, the set of input features into a set of tensors with a common feature dimension, wherein the set of embedding layers includes a linear layer for each input of the set of input features.
15 . The system of claim 14 , wherein inputting the set of input features into the transformer neural network further includes:
the transformer encoder layers receiving the mapped set of tensors with common feature dimension from the set of embedding layers; and
computing, using the transformer encoder layers, self-attention along temporal and agent dimensions to the mapped set of tensors with common feature dimensions to generate transformer encoder layer embeddings.
16 . The system of claim 15 , wherein inputting the set of input features into the transformer neural network further includes:
mapping, using the fully connected layers, the transformer encoder layer embeddings into tensors corresponding to target metrics.
17 . The system of claim 15 , wherein creating a set of inputs features occurs automatically upon detection of the sporting event being scheduled.
18 . The system of claim 15 , wherein the target metric predictions include: goals, assists, shots, shots on target, passes, fouls, yellow cards, red cards and/or minutes for one or more of the players.
19 . A non-transitory computer readable medium configured to store processor-readable instructions, wherein when executed by a processor, the instructions perform operations comprising:
receiving a set of input features, the set of input features representing a set of players and teams within a match, each input within the set of input features being represented by a tensor;
inputting the set of input features into a transformer neural network, the transformer neural network including:
a set of embedding layers;
transformer encoder layers; and
fully connected layers; and
generating, using the transformer neural network, a set of target metric predictions for the set of players and teams within the match.
20 . The non-transitory computer readable medium of claim 19 , wherein the target metric predictions include: goals, assists, shots, shots on target, passes, fouls, yellow cards, red cards and/or minutes for one or more of the players.