def add_rpm_to_player_table(year):
if int(year) < 2014:
print("No RPM data available prior to 2013-14 season")
return
year_string = construct_year_string(year)
rpm_filename = "{}/player_rpms_{}_rpms.csv".format(year_string, year)
with open(rpm_filename, "r") as csvfile:
data = csv.reader(csvfile)
for index, ele in enumerate(data):
# skip header row
if index == 0:
continue
team_codes = ele[2].split("/")
player_name = ele[0]
if not player_name:
continue
player_stub_name = common_utils.player_to_stub_name(player_name)
for code in team_codes:
if code in common_utils.team_codes_to_bball_ref_codes:
code = common_utils.team_codes_to_bball_ref_codes[code]
player_name = common_utils.nba_com_player_name_to_bball_ref_player_name(
player_name, code, year)
player_stub_name = common_utils.player_to_stub_name(
player_name)
rpm_update = {
"orpm": float(ele[5]),
"drpm": float(ele[6]),
"rpm": float(ele[7])
}
db.players.update_one(
{
'player_index.name_stub': player_stub_name,
'player_index.team': code,
'player_index.season': year
}, {'$set': rpm_update}, False)
print("Added RPM for {}".format(player_name))
def import_players_into_mongo(year):
year_string = construct_year_string(year)
print('creating index')
db.players.create_index([('player_index.name_stub', pymongo.ASCENDING),
('player_index.season', pymongo.ASCENDING),
('player_index.team', pymongo.ASCENDING)],
unique=True)
pprint(db.players.index_information())
print('index created (or already existed)')
with open("{}/players_advanced_{}.json".format(year_string, year),
"r") as playerfile:
player_json = json.load(playerfile)
headers = player_json.pop(0)
headers.pop(1)
headers.append(("Season", "season", None))
for index, player_row in enumerate(player_json):
player_row.append(year)
mongo_row = {}
for index, element in enumerate(player_row):
try:
mongo_row[headers[index][1]] = float(player_row[index])
except ValueError as e:
mongo_row[headers[index][1]] = player_row[index]
mongo_row["player"] = mongo_row["player"].replace("*", "")
stub_name = common_utils.player_to_stub_name(mongo_row["player"])
print("import {}_{}_{}".format(stub_name, mongo_row["team_id"],
year))
db.players.update_one(
{
'player_index.name_stub': stub_name,
'player_index.team': mongo_row["team_id"].strip(),
'player_index.season': year
}, {'$set': mongo_row}, True)
def store_player_and_possession_data_for_matrix(year, season_type):
player_info = {}
year_string = common_utils.construct_year_string(year)
year_as_int = int(year)
possessions = db.possessions[season_type][year_string].find().limit(limit)
count = 0
player_index = 0
num_possessions = possessions.count()
games_data = db.seasons.find_one({"year_string": year_string})["{}_games_data".format(season_type)]
#print(games_data)
# player info should already be calculated, this is to get index
if season_type == "playoffs":
num_players = len(db.seasons.find_one({ "year_string": year_string })["player_info"])
for possession in possessions:
#print(possession)
count += 1
if count % 20000 == 0:
print("Poss count {}/{}".format(count, num_possessions))
lineups_with_team_names = convert_lineups_to_player_team_strings(possession, games_data)
home_or_away = ["home_lineup", "away_lineup"]
for lineup_type in home_or_away:
for player_team in lineups_with_team_names[lineup_type]:
player_name, team_code = player_team.split("_")
stub_name = common_utils.player_to_stub_name(player_name)
#print(player_name, team_code)
if player_name == "None":
continue
if player_team not in player_info:
if team_code == "WAS" and year_as_int < 1998:
team_code = "WSB"
# found the player in bball ref database
if common_utils.player_exists(stub_name, team_code, year_as_int):
player_info[player_team] = {
"index": player_index,
"possessions": 0,
"stub_name" : stub_name
}
player_index += 1
# try to resolve name
else:
print("{}_{}_{} not found.".format(stub_name, team_code, year_as_int))
if season_type == "playoffs":
print("\n{}_{}_{} played in playoffs and not in regular season.".format(stub_name, team_code, year_as_int))
print("[{}][{}]: {}".format(
possession["possession_metadata"]["gid"],
possession["possession_metadata"]["event_num"],
possession["possession_metadata"]["message"]
))
print("\tHome", lineups_with_team_names["home_lineup"])
print("\tAway", lineups_with_team_names["away_lineup"])
db.seasons.find_one_and_update(
{ "year_string": year_string },
{ "$set" :
{ "player_info.{}".format(player_team):
{
"index": num_players,
"possessions": 0,
"stub_name": stub_name
}
}
}
)
num_players += 1
# player was already seen in a lineup
else:
player_info[player_team]["possessions"] += 1
for player_team in player_info:
possession_number = player_info[player_team]["possessions"]
player_name, team_code = player_team.split("_")
#print(player_info[player_team]["stub_name"], team_code)
db.players.update_one(
filter = {
"player_index.name_stub": player_info[player_team]["stub_name"],
"player_index.team": team_code,
"player_index.season": year_as_int
},
update = {
"$set": {
"{}_possessions".format(season_type): possession_number
}
},
upsert=True
)
if season_type == "regular_season":
db.seasons.find_one_and_update(
{ "year_string": year_string },
{ "$set" : { "player_info": player_info } }
)
def calculate_rapm(year, X, Y, metric="rapm", season_type="regular_season"):
year_as_int = int(year)
year_string = common_utils.construct_year_string(year)
print('fitting model...')
timestamp = time.perf_counter()
multiplier = 1
if metric == "apm":
clf = linear_model.LinearRegression()
multiplier = 10
elif metric == "rapm":
clf = linear_model.Ridge(alpha=2900)
multiplier = 100
elif metric == "rapm_enet_cv":
clf = linear_model.ElasticNetCV(l1_ratio=.9,cv=3)
elif metric == "rapm_bayes_ridge":
clf = linear_model.BayesianRidge()
elif metric == "rapm_cv":
clf = linear_model.RidgeCV(alphas=list(range(2600,3000,50)), cv=5)
multiplier = 100
else:
raise RuntimeError("{} not recognized".format(metric))
clf.fit(X, Y)
if metric == "rapm_enet_cv" or metric == "rapm_cv":
print("alpha chosen: {}".format(clf.alpha_))
if metric == "rapm_enet_cv":
print("l1_ratio chosen: {}".format(clf.l1_ratio_))
print("time took {} seconds".format(time.perf_counter()-timestamp))
all_players_dict = db.seasons.find_one({"year_string": year_string})["player_info"]
print("coefficients", clf.coef_.shape)
print("num of players", len(all_players_dict))
rapm = list(zip(all_players_dict, clf.coef_*multiplier, clf.coef_[len(all_players_dict):]*-1*multiplier))
print(all_players_dict)
print(len(rapm))
for player, opm, dpm in rapm:
player_name, team_code = player.split("_")
data = {
'player': player_name,
'team': team_code,
'o' + metric: opm,
'd' + metric: dpm,
metric: opm + dpm
}
stub_name = common_utils.player_to_stub_name(data["player"])
team_code = data["team"]
year_as_int = int(year)
print ("updating {}_{}_{}".format(stub_name, team_code, year_as_int))
db.players.update_one(
{
"player_index.name_stub": stub_name,
"player_index.team": team_code,
"player_index.season": year_as_int
},
{
"$set": {
"o" + metric + "_" + season_type: data["o" + metric],
"d" + metric + "_" + season_type: data["d" + metric],
metric + "_" + season_type: data[metric]
}
}
)
print("Sorted & filtered {} for {}".format(metric, year_string))
poss_threshold = 1000
sorted_opm = db.players.find(
filter = {"player_index.season": year_as_int, "possessions": {"$gt": poss_threshold}},
projection = {"player": 1, metric + "_" + season_type: 1},
sort = [(metric + "_" + season_type, -1)]
)
rankings_to_print = 50
for player in sorted_opm:
if "player" in player:
print("{:20} {:>10}".format(player["player"], player[metric + "_" + season_type]))
rankings_to_print += -1
if rankings_to_print == 0:
break
return clf
def get_and_parse_event_list_for_lineups(event_list, team_info, year):
"""
Take rowSet of events from API call and return a list of equal size with the 5 man lineup on the floor for each team for each event
Example output format:
[
{
"Lakers": ["Lonzo Ball", "Brandon Ingram", "Corey Brewer", "Julius Randle", "Brook Lopez"],
"Warriors": ["Stephen Curry", "Kevin Durant", "Klay Thompson", "Draymond Green", "Andre Iguodala"],
"event_num": 2
}, ...
]
"""
player_ids = {}
year_as_int = int(year)
home_team = team_info["home_team"]
away_team = team_info["away_team"]
lineups = {team_info["home_team"]: set(), team_info["away_team"]: set()}
event_with_lineups_list = []
current_quarter = -1
for index, event in enumerate(event_list):
#print(event_with_lineups_list)
# reset lineups when quarter changes
if current_quarter != event[event_field["PERIOD"]]:
current_quarter = event[event_field["PERIOD"]]
lineups[team_info["home_team"]] = set()
lineups[team_info["away_team"]] = set()
quarter_start_index = index
# this piece of code is for a single empty event toward the end of this game:
# http://stats.nba.com/game/0020000883/playbyplay/ that causes problems otherwise
if event[event_field["EVENTMSGTYPE"]] == event_type[
"SUB"] and get_message(event) is None:
pass
# substitution event
# player 1 is being substituted out
# player 2 is coming in
elif event[event_field["EVENTMSGTYPE"]] == event_type["SUB"]:
#print_event(event)
team = common_utils.convert_nba_espn_team_codes_to_bball_ref(
event[event_field["PLAYER1_TEAM_ABBREVIATION"]], year_as_int)
try:
player_to_sub_out = common_utils.nba_com_player_name_to_bball_ref_player_name(
event[event_field["PLAYER1_NAME"]], team, year_as_int)
except RuntimeError as e:
print_event(event)
try:
player_to_sub_in = common_utils.nba_com_player_name_to_bball_ref_player_name(
event[event_field["PLAYER2_NAME"]], team, year_as_int)
except RuntimeError as e:
print_event(event)
if (player_to_sub_out, team, year_as_int) not in player_ids:
player_ids[(player_to_sub_out, team,
year_as_int)] = event[event_field["PLAYER1_ID"]]
if (player_to_sub_in, team, year_as_int) not in player_ids:
player_ids[(player_to_sub_in, team,
year_as_int)] = event[event_field["PLAYER2_ID"]]
if player_to_sub_out not in lineups[team]:
# backfill
for e in event_with_lineups_list[quarter_start_index:]:
e[team].add(player_to_sub_out)
lineups[team].discard(player_to_sub_out)
lineups[team].add(player_to_sub_in)
event_with_lineups_list.append({
team_info["home_team"]:
lineups[team_info["home_team"]].copy(),
team_info["away_team"]:
lineups[team_info["away_team"]].copy(),
"event_num":
event[event_field["EVENTNUM"]],
})
# player1, player2, player3
if event[event_field["EVENTMSGTYPE"]] != event_type["SUB"]:
f = [
event_field['PLAYER1_NAME'], event_field['PLAYER2_NAME'],
event_field['PLAYER3_NAME']
]
for p_index in f:
# team code is always an offset of 4 from player name
if event[p_index] and event[p_index + 4]:
team_code = common_utils.convert_nba_espn_team_codes_to_bball_ref(
event[p_index + 4], year_as_int)
try:
player_name = common_utils.nba_com_player_name_to_bball_ref_player_name(
event[p_index], team_code, year_as_int)
except RuntimeError as e:
print(e)
if (player_name, team_code, year_as_int) not in player_ids:
# player id is offset -1 from player name
player_ids[(player_name, team_code,
year_as_int)] = event[p_index - 1]
if len(lineups[team_code]
) < 5 and player_name not in lineups[team_code]:
lineups[team_code].add(player_name)
# backfill
for e in event_with_lineups_list[quarter_start_index:]:
e[team_code].add(player_name)
for (t_player_name, t_team_code, t_year) in player_ids:
stub_name = common_utils.player_to_stub_name(t_player_name)
player = db.players.update_one(
{
"player_index.name_stub": stub_name,
"player_index.team": t_team_code,
"player_index.season": t_year
}, {
"$set": {
"nba_com_id":
player_ids[(t_player_name, t_team_code, t_year)]
}
})
return event_with_lineups_list
def calculate_weighted_average(year):
year_as_int = int(year)
metrics_averages = {
"per": 15.0,
"bpm": 0.0,
"ws_per_48": 0.100,
"rapm": 0.0,
"apm": 0.0
}
if int(year) >= 2014:
metrics_averages["rpm"] = 0.0
year_as_int = int(year)
year_string = common_utils.construct_year_string(year)
teams = db.teams.find({"season": year})
for team in teams:
team_code = team["team_code"]
# get the players to loop through
players = db.players.find({
"player_index.season": year_as_int,
"player_index.team": team_code
})
players = list(players)
if len(players) == 0 and team_code in common_utils.team_renames:
players = db.players.find({
"player_index.season":
year_as_int,
"player_index.team":
common_utils.team_renames[team_code]
})
players = list(players)
if len(players) == 0:
print("could not find {}_{}".format(year_to_evaluate,
team["team_index"]["team_id"]))
continue
for year_step in range(4):
timestamp = time.perf_counter()
year_to_evaluate = year - year_step
total_possessions = 0
weighted_average = 0
for metric in metrics_averages:
print("{} for {} {}\n".format(metric, year_to_evaluate,
team_code))
if year_to_evaluate < 1997:
print("no data available for prior to 1997")
continue
for player in players:
# original measure was 250 minutes
if "possessions" in player and "player" in player:
stub_name = common_utils.player_to_stub_name(
player["player"])
prev_season_player_lookup = db.players.find({
"player_index.season":
year_to_evaluate,
"player_index.name_stub":
stub_name
})
if not player_exists(stub_name, year_to_evaluate):
if player_minus_jr_exists(stub_name,
year_to_evaluate):
stub_name = stub_name[:len(stub_name) - 2]
print("substracted suffix, found {}".format(
stub_name))
if stub_name in common_utils.nba_com_stubs_to_bball_ref_stubs:
stub_name = common_utils.nba_com_stubs_to_bball_ref_stubs[
stub_name]
# if player still doesn't exist, then idk
if not player_exists(stub_name, year_to_evaluate):
print(
"{}_{} not found in scraped bball_ref players page"
.format(stub_name, year_to_evaluate))
else:
prev_season_player_lookup = db.players.find({
"player_index.season":
year_to_evaluate,
"player_index.name_stub":
stub_name,
"player_index.team":
team_code
})
metric_value = 0
prev_poss_count = 0
# calculate metric value (done in this way because to combine entries for players who were traded)
for prev_player_season in prev_season_player_lookup:
if "possessions" in prev_player_season:
try:
prev_poss_count += prev_player_season[
"possessions"]
#print("\t{} - {}: {} Poss {}".format(player["player"], metric, player[metric], player["possessions"]))
metric_value += float(
prev_player_season[metric]
) * prev_player_season["possessions"]
except:
print("error trying to convert to float ",
metric, prev_player_season)
else:
metric_value = metrics_averages[metric]
if prev_poss_count > 750:
metric_value = metric_value / prev_poss_count
else:
metric_value = metrics_averages[metric]
try:
total_possessions += player["possessions"]
weighted_average += player[
"possessions"] * metric_value
except:
print(player)
else:
player_name = player["player"] if player.get(
"player") else "unknown_player_name"
team_code = player["team_id"] if player.get(
"team_id") else "unknown_team"
player_key = "{}_{}_{}".format(player_name, year,
team_code)
players_without_possessions.add(player_key)
if total_possessions == 0:
print(players)
continue
if total_possessions > 0:
weighted_average = weighted_average / total_possessions
print("{}_{} has a weighted average of {} {}\n".format(
team_code, year_to_evaluate, weighted_average, metric))
db.teams.update_one(
{
"team_index.season": year,
"team_index.team_id": team_code
}, {
"$set": {
"Y-{}_{}".format(year_step, metric):
weighted_average
}
})
total_possessions = 0
weighted_average = 0
print("time took {} seconds for one year step".format(
time.perf_counter() - timestamp))
pprint(players_without_possessions)
