dolly-the-sheep/NFL_game_footage

If you, like me, are interested in using AI to win sports, reach out to me to collaborate on this project.

The purpose of this dataset would be to accomplish numerous tasks including but no limited to creating an AI nfl head coach, general manager, refs, predicting outcomes... Numerous machine-learning steps need to be accomplished on the way. Those include the following: 1: Take NFL game footage (all 22) and split it into plays 2: Classify each play according to many categorical groupings including but not limited to:

• Play type/ play name
• Play result (literal)
• Play formations
• Play value (quantitative)
• Individual player actions
• Individual player assignments/reads
• Individual player success on assignemnts
• Team success on play
• Flags on play

3. A realistic style NFL simulation can be developed and machine learning can be used to coorelate footage to simulated plays.

• Simulation should include individual player assignments per play (such that an NFL head coach could verbally describe to their players)
• Simulation should not treat all players (or all players of same position) as identical, rather it should create quauntiative descriptions for each player including levels of uncertainty

4. Once a simulation that corresponds well with actual gameplay and coach-style-descriptions of play, reinforcement learning can be used to develop advanced gameplay strategies (like Chess and GO) which can be translated into coaching techniques
5. Player value should be easier to determine via this simulation.

Steps to take:

1. Gather relevant footage (likely need to buy it from NFL)
2. Design algorithms to expand the dataset by including the information that can be gained from the footage. (Probably using human annotation which can cause problems) (Humans might not be great at adequately valuing play outcomes)
3. Use expanded dataset to accopmlish tasks aboves.