System and method for multi-task learning
A method of generating a multi-modal prediction is disclosed herein. A computing system retrieves event data from a data store. The event data includes information for a plurality of events across a plurality of seasons. Computing system generates a predictive model using a mixture density network, by generating an input vector from the event data learning, by the mixture density network, a plurality of values associated with a next play following each play in the event data. The mixture density network is trained to output the plurality of values near simultaneously. Computing system receives a set of event data directed to an event in a match. The set of event data includes information directed to at least playing surface position and current score. Computing system generates, via the predictive model, a plurality of values associated with a next event following the event based on the set of event data.
1. A method of generating a multi-modal prediction, comprising:
retrieving, by a computing system, event data from a data store, the event data comprising information for a plurality of events across a plurality of seasons;
generating, by the computing system, a predictive model using a mixture density network, by:
generating an input vector from the event data, the input vector comprising one or more parameters associated with each play in the event data; and
learning, by the mixture density network, a plurality of values associated with a next play following each play in the event data, wherein the mixture density network is trained to output the plurality of values near simultaneously;
receiving, by the computing system, a set of data directed to an event in a match, the set of data comprising information directed to at least playing surface position and current score; and
generating, by the computing system via the predictive model, a plurality of values associated with a next event following the event based on the set of data, wherein the plurality of values is determined near simultaneously.
2. The method of claim 1 , wherein generating the input vector from the event data, comprises:
for each play in the event data, segmenting data corresponding thereto.
3. The method of claim 1 , wherein the event data for each play in the event data comprises categorical features and continuous features.
4. The method of claim 3 , wherein generating the input vector from the event data, comprises:
passing the categorical features through a respective embedding layer to create a dense representation of each categorical feature.
5. The method of claim 4 , further comprising:
concatenating the dense representation of each categorical feature with the continuous features to generate the input vector.
6. The method of claim 4 , wherein the input vector comprises a dense representation of playing surface position data and a raw representation of playing surface position data.
7. The method of claim 1 , wherein generating, by the computing system via the predictive model, the plurality of values associated with the next event following the event based on the set of data comprises:
generating an output vector comprising one or more of expected meters gained, expected play selection, likelihood of scoring on the next event, likelihood of scoring during a sequence of events comprising the next event, likelihood of winning the event, a scoreline prediction, expected goal, expected shot, expected foul/penalty, expected corner, win probability, and final score line, expected ace, expected winning of point, expected break, win probability, final score, expected number of points being scored by a specific player, expected number of rebounds per specific player, win probability, and final score prediction, depending on which sport is identified.
8. A system for generating a multi-modal prediction, comprising:
a processor; and
a memory having programming instructions stored thereon, which, when executed by the processor, performs one or more operations, comprising: retrieving, by a computing system, event data from a data store, the event data comprising information for a plurality of events across a plurality of seasons;
generating a predictive model using a mixture density network, by:
generating an input vector from the event data, the input vector comprising one or more parameters associated with each play in the event data; and
learning, by the mixture density network, a plurality of values associated with a next play following each play in the event data, wherein the mixture density network is trained to output the plurality of values near simultaneously;
receiving a set of data directed to an event in a match, the set of data comprising information directed to at least playing surface position and current score; and
generating, via the predictive model, a plurality of values associated with a next event following the event based on the set of data, wherein the plurality of values are determined near simultaneously.
9. The system of claim 8 , wherein generating the input vector from the event data, comprises:
for each play in the event data, segmenting data corresponding thereto.
10. The system of claim 8 , wherein the event data for each play in the event data comprises categorical features and continuous features.
11. The system of claim 10 , wherein generating the input vector from the event data, comprises:
passing the categorical features through a respective embedding layer to create a dense representation of each categorical feature.
12. The system of claim 11 , further comprising:
concatenating the dense representation of each categorical feature with the continuous features to generate the input vector.
13. The system of claim 12 , wherein the input vector comprises a dense representation of playing surface position data and a raw representation of playing surface position data.
14. The system of claim 8 , wherein generating, via the predictive model, the plurality of values associated with the next event following the event based on the set of data comprises:
generating an output vector comprising one or more of expected meters gained, expected play selection, likelihood of scoring on the next event, likelihood of scoring during a sequence of events comprising the next event, likelihood of winning the event, a scoreline prediction, expected goal, expected shot, expected foul/penalty, expected corner, win probability, and final score line, expected ace, expected winning of point, expected break, win probability, final score, expected number of points being scored by a specific player, expected number of rebounds per specific player, win probability, and final score prediction, depending on which sport is identified.
15. A non-transitory computer readable medium including one or more sequences of instructions that, when executed by one or more processors, causes a computing system to perform operations comprising:
retrieving, by a computing system, event data from a data store, the event data comprising information for a plurality of events across a plurality of seasons;
generating, by the computing system, a predictive model using a mixture density network, by:
generating an input vector from the event data, the input vector comprising one or more parameters associated with each play in the event data; and
learning, by the mixture density network, a plurality of values associated with a next play following each play in the event data, wherein the mixture density network is trained to output the plurality of values near simultaneously;
receiving, by the computing system, a set of data directed to an event in a match, the set of data comprising information directed to at least playing surface position and current score; and
generating, by the computing system via the predictive model, a plurality of values associated with a next event following the event based on the set of data, wherein the plurality of values are determined near simultaneously.
16. The non-transitory computer readable medium of claim 15 , wherein generating the input vector from the event data, comprises:
for each play in the event data, segmenting data corresponding thereto.
17. The non-transitory computer readable medium of claim 15 , wherein the event data for each play in the event data comprises categorical features and continuous features.
18. The non-transitory computer readable medium of claim 17 , wherein generating the input vector from the event data, comprises:
passing the categorical features through a respective embedding layer to create a dense representation of each categorical feature.
19. The non-transitory computer readable medium of claim 18 , further comprising:
concatenating the dense representation of each categorical feature with the continuous features to generate the input vector.
20. The non-transitory computer readable medium of claim 15 , wherein generating, by the computing system via the predictive model, the plurality of values associated with the next event following the event based on the set of data comprises:
generating an output vector comprising one or more of expected meters gained, expected play selection, likelihood of scoring on the next event, likelihood of scoring during a sequence of events comprising the next event, likelihood of winning the event, a scoreline prediction, expected goal, expected shot, expected foul/penalty, expected corner, win probability, and final score line, expected ace, expected winning of point, expected break, win probability, final score, expected number of points being scored by a specific player, expected number of rebounds per specific player, win probability, and final score prediction, depending on which sport is identified.