This project was a first-round sprint project for the 2019 NBA hackathon. Given official NBA play-by-play data, we needed to calculate the offensive and defensive ratings for players per-game.

We were given data to 82 games in data/Play_by_Play.txt to process, but our script is generalizable to any game.

We calculated the offensive rating for each player with this formula:

Off-rating = (points scored by team while on court) / (offensive possessions involved in) * 100

and we calculated the defensive rating with this formula:

Def-rating = (points allowed by team while on court) / (defensive possessions involved in) * 100

Knowing this, we just need to keep track of:

1) points scored/allowed by the team when the player is on the court and

2) number of possessions the player is part of.

Note that both of these items are unique to the player.

Don’t be like pop, count your numbers right.

We created 3 classes to help our endeavors: Game, Team and Player.

Player instances will record the 2 items mentioned above. Team instances will keep track of all players and the active players (on the court) and increment their points and possession count as appropriate. Game instances ingests all the data needed for that one game and modifies its Team instances to reflect substitutions, points scored and more.

During initialization, a Game instance instantiates 2 Team instances, which each hold 15 Player instances.

There existed many events in the Event_Codes data given to us. We used our basketball knowledge to determine which events would be relevant to our offensive and defensive rating calculations. Here is how we treated each event programmatically:

It becomes slightly problematic when a substitution happens after a foul but before a free throw. This is because the NBA officially associates that free throw point to the active players on the court before the foul occurred. Logically, those 10 players were responsible for the foul happening, and not the players subbed in.

To accomodate this, we kept track of 2 lists of active players once a foul happened. After a foul ends (last free throw finishes and shot clock continues), we would continue just keeping track of 1 list of active players.

active_1 represented the active players throughout the game except when free throws were happening. If free throws were happening, active_1 would list the players on the court before any substitutions happen and active_2 would list the players on the court factoring in the substitutions (if there were substitutions). Anytime free throws ended, active_2 would be moved into active_1 and active_2 would be deleted. active_1 would continue to keep track of the active players on the court.

We would always add points / possessions to the players in active­_1. However, there is one special case in which we do add points / possessions to the players in active_2 : when technical free throws happen in between free throws for another type of foul (normal foul, flagrant foul, clear path foul). In this case, we believe the technical foul arose directly due to the players on court, not the players on court at the time of the other foul. Hence, we would add technical free throw points to active_2 in this case.

We also understood that the definition of “a new possession” had major implications on our analysis. Here is how we handled the definitions possessions for fouls:

For Flagrant and Clear Path fouls, we counted the completion of their free throws as the end of a possession. The NBA Rulebook tells us that Flagrant and Clear Path fouls result in free throws and possession, so we counted completing their free throws as a separate possession.

For technical fouls, we did not count them as a separate possession. We believe that technical free throws are a result of unsportsmanlike conduct but should not affect the possession flow of the game. Hence, we would have the completion of technical free throws not count towards a possession.

We found 3 games in the play-by-play data that had discrepancies with the other 79 games. There were no jump ball events for these 3 games at the start of the game or start of overtime.

890600997c60cca3e41236355c65642e

d508da19217d10b9a4566ef06184b4fd

a466c76e072fd634f6d4a8938fb63caa

For these games we manually inspected the data and deduced who got possession from the jump balls.