Injuries

[t-dmac]

Does the model account for injuries ?

I ask because for example, today's Lakers and Pacers game, it's predicting Lakers with a small chance of winning. That would make sense with Bron and Luka, but they're actually missing Bron, Luka and Reaves and one more placer is also questionable.

Thoughts ?

[nealmick]

The models uses a combination of stats consisting of team-based averages such as points, rebounds, and three-point shots. The model also uses current odds data to help make informed decisions. Injuries are sort of baked into these stats - if a team's top player is injured, then that team's average points and other stats will suffer. Also, the odds will be impacted by injury news and is passed to the model; the model is aware of current market value. The base project does not include specific player stats. I do have a branch that adds last game's data that contains all of the stats from each player's last game such as fouls, steals, points, and so forth. With last game stats, the model is able to see if a player is injured via their stats from the previous game, and it also shows more granular and current momentum-type data of the recent performance, giving the model more info to calibrate.

There is a way to train a model with, let's say, the top 5 players' season averages, which can be determined by most points or playtime... and then remove the players and replace them with the next best whenever there is an injury. This allows us to run inference with different combinations of rosters, as sometimes it's unclear if a player will be benched. This does require historical injury data to work best; however, when training, you can assume the roster will be the same as the last game, which is mostly true and will pick up most injuries... and then you can make a final decision based on doctors' reports when it comes time to predict future games, what roster combinations make sense, and how they potentially could affect game results.

[mendez97cr]

email me bro.

[t-dmac]

Thank you, perfectly explained, you're a genius. 〽️ac

…

On Feb 8, 2025 at 9:23 PM -0500, Neal ***@***.***>, wrote: The models uses a combination of stats consisting of team-based averages such as points, rebounds, and three-point shots. The model also uses current odds data to help make informed decisions. Injuries are sort of baked into these stats - if a team's top player is injured, then that team's average points and other stats will suffer. Also, the odds will be impacted by injury news and is passed to the model; the model is aware of current market value. The base project does not include specific player stats. I do have a branch that adds last game's data that contains all of the stats from each player's last game such as fouls, steals, points, and so forth. With last game stats, the model is able to see if a player is injured via their stats from the previous game, and it also shows more granular and current momentum-type data of the recent performance, giving the model more info to calibrate. There is a way to train a model with, let's say, the top 5 players' season averages, which can be determined by most points or playtime... and then remove the players and replace them with the next best whenever there is an injury. This allows us to run inference with different combinations of rosters, as sometimes it's unclear if a player will be benched. This does require historical injury data to work best; however, when training, you can assume the roster will be the same as the last game, which is mostly true and will pick up most injuries... and then you can make a final decision based on doctors' reports when it comes time to predict future games, what roster combinations make sense, and how they potentially could affect game results. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>