This will be a repository where I store all my scripts and tests for compiling and calculating NBA game data from play by play dataframe objects.
The main hook of the nba_parser will the the PbP class that will take a play
by play Pandas dataframe either as the direct output from my nba_scraper
or as a Pandas dataframe created from the csv output of the nba_scraper saved
to file.
Player stats can be calculated from a play by play dataframe with just a few lines of code.
import nba_scraper.nba_scraper as ns
import nba_parser as npar
game_df = ns.scrape_game([20700233])
pbp = npar.PbP(game_df)
player_stats = pbp.playerbygamestats()
#can also derive single possessions for RAPM calculations
rapm_shifts = pbp.rapm_possessions()Which produces a dataframe containing the stats of field goals made, field goals attempted, three points made, three points attempted, free throws made, free throws attempted, steals, turnovers, blocks, personal fouls, minutes played(toc), offensive rebounds, possessions and defensive rebounds.
Team stats are called very similar to player stats.
import nba_scraper.nba_scraper as ns
import nba_parser as npar
game_df = ns.scrape_game([20700233])
pbp = npar.PbP(game_df)
team_stats = pbp.teambygamestats()The team stats that will be calculation are field goals made, field goals attempted, three points made, three points attempted, free throws made, free throws attempted, steals, turnovers, blocks, personal fouls, minutes played(toc), offensive rebounds, possessions, home team, winning team, fouls drawn, shots blocked, total points for, total points against, and defensive rebounds.
I've grouped together other stat calculations that work better with larger sample sizes.
This class takes a list of outputs from PbP.teambygamestats() but really it could take a
list of dataframes that are the same structure as that method output. Here's an example
of how it could work in conjunction with nba_scraper. I suggest writing the pbp returns to file and then importing them as the nba_scraper could time out due to the NBA api timing out from being hit too many times.
import nba_scraper.nba_scraper as ns
import nba_parser as npar
tbg_dfs = []
for game_id in range(20700001, 20700010):
game_df = ns.scrape_game([game_id])
pbp = np.PbP(game_df)
team_stats = pbp.teambygamestats()
tbg_dfs.append(team_stats)
team_totals = npar.TeamTotals(tbg_dfs)
#produce a dataframe of eFG%, TS%, TOV%, OREB%, FT/FGA, Opponent eFG%,
#Opponent TOV%, DREB%, Opponent FT/FGA, along with summing the other
#stats produced by the teambygamestats() method to allow further
#calculations
team_adv_stats = team_totals.team_advanced_stats()
#to calculate a RAPM regression for teams use this method
team_rapm_df = team_totals.team_rapm_results()Like with TeamTotals i've grouped player stat calculations that work better with a larger sample size into its own class. A lot of the hooks are similar except for the RAPM calculation which is a static method due to the time to calculate player RAPM shifts is much longer than team shifts so its best to have them precalculated before attempting a RAPM regression to reduce time.
import nba_scraper.nba_scraper as ns
import nba_parser as npar
pbg_dfs = []
pbp_objects = []
for game_id in range(20700001, 2070010):
game_df = ns.scrape_game([game_id])
pbp = np.PbP(game_df)
pbp_objects.append(pbp)
player_stats = pbp.playerbygamestats()
pbg_dfs.append(player_stats)
player_totals = npar.PlayerTotals(pbg_dfs)
#produce a dataframe of eFG%, TS%, TOV%, OREB%, AST%, DREB%,
#STL%, BLK%, USG%, along with summing the other
#stats produced by the playerbygamestats() method to allow further
#calculations
player_adv_stats = player_totals.player_advanced_stats()
#to calculate a RAPM regression for players first have to calculate
#RAPM possessions from the list of PbP objects we collected above
rapm_possession = pd.concat([x.rapm_possessions() for x in pbp_objects])
player_rapm_df = npar.PlayerTotals.player_rapm_results(rapm_possession)