def _team_from_xml_elm(team_elm, f7_root) -> Team:
# This should not happen here
team_name, team_players = _parse_team_players(f7_root,
team_elm.attrib["TeamRef"])
team_id = team_elm.attrib["TeamRef"].lstrip("t")
team = Team(
team_id=str(team_id),
name=team_name,
ground=Ground.HOME
if team_elm.attrib["Side"] == "Home" else Ground.AWAY,
)
team.players = [
Player(
player_id=player_elm.attrib["PlayerRef"].lstrip("p"),
team=team,
jersey_no=int(player_elm.attrib["ShirtNumber"]),
first_name=team_players[
player_elm.attrib["PlayerRef"]]["first_name"],
last_name=team_players[player_elm.attrib["PlayerRef"]]
["last_name"],
position=Position(
position_id=player_elm.attrib["Formation_Place"],
name=player_elm.attrib["Position"],
coordinates=None,
),
) for player_elm in team_elm.find("PlayerLineUp").iterchildren(
"MatchPlayer")
]
return team
def __create_anon_player(cls, teams, frame_record):
"""
creates a Player object for a track_id'ed player with known team membership but unknown identity.
Args:
frame_record (dict): dictionary containing 'x', 'y', 'track_id' and 'group_name'
Returns:
kloppy.domain.models.common.Player
"""
track_id = frame_record.get("track_id", None)
group_name = frame_record.get("group_name", None)
if group_name == "home team":
team = teams[0]
elif group_name == "away team":
team = teams[1]
else:
raise ValueError(
f"anonymous player with track_id `{track_id}` does not have a specified group_name."
)
return Player(
player_id=f"{team.ground}_anon_{track_id}",
team=team,
jersey_no=None,
name=f"Anon_{track_id}",
first_name="Anon",
last_name=track_id,
starting=None,
position=None,
attributes={},
)
def _team_from_xml_elm(team_elm) -> Team:
team = Team(
team_id=team_elm.attrib["TeamId"],
name=team_elm.attrib["TeamName"],
ground=Ground.HOME
if team_elm.attrib["Role"] == "home"
else Ground.AWAY,
)
team.players = [
Player(
player_id=player_elm.attrib["PersonId"],
team=team,
jersey_no=int(player_elm.attrib["ShirtNumber"]),
name=player_elm.attrib["Shortname"],
first_name=player_elm.attrib["FirstName"],
last_name=player_elm.attrib["LastName"],
position=Position(
position_id=None,
name=player_elm.attrib["PlayingPosition"],
coordinates=None,
)
if "PlayingPosition" in player_elm.attrib
else None,
starting=player_elm.attrib["Starting"] == "true",
)
for player_elm in team_elm.Players.iterchildren("Player")
]
return team
def _get_tracking_dataset(self):
home_team = Team(team_id="home", name="home", ground=Ground.HOME)
away_team = Team(team_id="away", name="away", ground=Ground.AWAY)
teams = [home_team, away_team]
periods = [
Period(
id=1,
start_timestamp=0.0,
end_timestamp=10.0,
attacking_direction=AttackingDirection.HOME_AWAY,
),
Period(
id=2,
start_timestamp=15.0,
end_timestamp=25.0,
attacking_direction=AttackingDirection.AWAY_HOME,
),
]
metadata = Metadata(
flags=~(DatasetFlag.BALL_OWNING_TEAM | DatasetFlag.BALL_STATE),
pitch_dimensions=PitchDimensions(x_dim=Dimension(0, 100),
y_dim=Dimension(-50, 50)),
orientation=Orientation.HOME_TEAM,
frame_rate=25,
periods=periods,
teams=teams,
score=None,
provider=None,
)
tracking_data = TrackingDataset(
metadata=metadata,
records=[
Frame(
frame_id=1,
timestamp=0.1,
ball_owning_team=None,
ball_state=None,
period=periods[0],
players_coordinates={},
ball_coordinates=Point(x=100, y=-50),
),
Frame(
frame_id=2,
timestamp=0.2,
ball_owning_team=None,
ball_state=None,
period=periods[0],
players_coordinates={
Player(team=home_team, player_id="home_1", jersey_no=1):
Point(x=15, y=35)
},
ball_coordinates=Point(x=0, y=50),
),
],
)
return tracking_data
def _frame_from_framedata(cls, teams, period, frame_data):
frame_id = frame_data["frameIdx"]
frame_timestamp = frame_data["gameClock"]
if frame_data["ball"]["xyz"]:
ball_x, ball_y, ball_z = frame_data["ball"]["xyz"]
ball_coordinates = Point3D(
float(ball_x), float(ball_y), float(ball_z)
)
else:
ball_coordinates = None
ball_state = BallState.ALIVE if frame_data["live"] else BallState.DEAD
ball_owning_team = (
teams[0] if frame_data["lastTouch"] == "home" else teams[1]
)
players_data = {}
for team, team_str in zip(teams, ["homePlayers", "awayPlayers"]):
for player_data in frame_data[team_str]:
jersey_no = player_data["number"]
x, y, _ = player_data["xyz"]
player = team.get_player_by_jersey_number(jersey_no)
if not player:
player = Player(
player_id=player_data["playerId"],
team=team,
jersey_no=int(jersey_no),
)
team.players.append(player)
players_data[player] = PlayerData(
coordinates=Point(float(x), float(y))
)
return Frame(
frame_id=frame_id,
timestamp=frame_timestamp,
ball_coordinates=ball_coordinates,
ball_state=ball_state,
ball_owning_team=ball_owning_team,
players_data=players_data,
period=period,
other_data={},
)
def _parse_team(raw_events, wyId: str, ground: Ground) -> Team:
team = Team(
team_id=wyId,
name=raw_events["teams"][wyId]["officialName"],
ground=ground,
)
team.players = [
Player(
player_id=str(player["playerId"]),
team=team,
jersey_no=None,
first_name=player["player"]["firstName"],
last_name=player["player"]["lastName"],
) for player in raw_events["players"][wyId]
]
return team
def deserialize(
self, inputs: Dict[str, Readable], options: Dict = None
) -> EventDataset:
"""
Deserialize StatsBomb event data into a `EventDataset`.
Parameters
----------
inputs : dict
input `event_data` should point to a `Readable` object containing
the 'json' formatted event data. input `lineup_data` should point
to a `Readable` object containing the 'json' formatted lineup data.
options : dict
Options for deserialization of the StatsBomb file. Possible options are
`event_types` (list of event types) to specify the event types that
should be returned. Valid types: "shot", "pass", "carry", "take_on" and
"generic". Generic is everything other than the first 4. Those events
are barely parsed. This type of event can be used to do the parsing
yourself.
Every event has a 'raw_event' attribute which contains the original
dictionary.
Returns
-------
dataset : EventDataset
Raises
------
See Also
--------
Examples
--------
>>> serializer = StatsBombSerializer()
>>> with open("events/12312312.json", "rb") as event_data, \
>>> open("lineups/123123123.json", "rb") as lineup_data:
>>>
>>> dataset = serializer.deserialize(
>>> inputs={
>>> 'event_data': event_data,
>>> 'lineup_data': lineup_data
>>> },
>>> options={
>>> 'event_types': ["pass", "take_on", "carry", "shot"]
>>> }
>>> )
"""
self.__validate_inputs(inputs)
if not options:
options = {}
with performance_logging("load data", logger=logger):
raw_events = json.load(inputs["event_data"])
home_lineup, away_lineup = json.load(inputs["lineup_data"])
(
shot_fidelity_version,
xy_fidelity_version,
) = _determine_xy_fidelity_versions(raw_events)
logger.info(
f"Determined Fidelity versions: shot v{shot_fidelity_version} / XY v{xy_fidelity_version}"
)
with performance_logging("parse data", logger=logger):
home_team = Team(
team_id=str(home_lineup["team_id"]),
name=home_lineup["team_name"],
ground=Ground.HOME,
)
home_team.players = [
Player(
player_id=str(player["player_id"]),
team=home_team,
name=player["player_name"],
jersey_no=int(player["jersey_number"]),
)
for player in home_lineup["lineup"]
]
away_team = Team(
team_id=str(away_lineup["team_id"]),
name=away_lineup["team_name"],
ground=Ground.AWAY,
)
away_team.players = [
Player(
player_id=str(player["player_id"]),
team=away_team,
name=player["player_name"],
jersey_no=int(player["jersey_number"]),
)
for player in away_lineup["lineup"]
]
teams = [home_team, away_team]
wanted_event_types = [
EventType[event_type.upper()]
for event_type in options.get("event_types", [])
]
periods = []
period = None
events = []
for raw_event in raw_events:
if raw_event["team"]["id"] == home_lineup["team_id"]:
team = teams[0]
elif raw_event["team"]["id"] == away_lineup["team_id"]:
team = teams[1]
else:
raise Exception(
f"Unknown team_id {raw_event['team']['id']}"
)
if (
raw_event["possession_team"]["id"]
== home_lineup["team_id"]
):
possession_team = teams[0]
elif (
raw_event["possession_team"]["id"]
== away_lineup["team_id"]
):
possession_team = teams[1]
else:
raise Exception(
f"Unknown possession_team_id: {raw_event['possession_team']}"
)
timestamp = parse_str_ts(raw_event["timestamp"])
period_id = int(raw_event["period"])
if not period or period.id != period_id:
period = Period(
id=period_id,
start_timestamp=(
timestamp
if not period
# period = [start, end], add millisecond to prevent overlapping
else timestamp + period.end_timestamp + 0.001
),
end_timestamp=None,
)
periods.append(period)
else:
period.end_timestamp = period.start_timestamp + timestamp
player = None
if "player" in raw_event:
player = team.get_player_by_id(raw_event["player"]["id"])
event_type = raw_event["type"]["id"]
if event_type == SB_EVENT_TYPE_SHOT:
fidelity_version = shot_fidelity_version
elif event_type in (
SB_EVENT_TYPE_CARRY,
SB_EVENT_TYPE_DRIBBLE,
SB_EVENT_TYPE_PASS,
):
fidelity_version = xy_fidelity_version
else:
# TODO: Uh ohhhh.. don't know which one to pick
fidelity_version = xy_fidelity_version
generic_event_kwargs = dict(
# from DataRecord
period=period,
timestamp=timestamp,
ball_owning_team=possession_team,
ball_state=BallState.ALIVE,
# from Event
event_id=raw_event["id"],
team=team,
player=player,
coordinates=(
_parse_coordinates(
raw_event.get("location"), fidelity_version
)
if "location" in raw_event
else None
),
raw_event=raw_event,
)
if event_type == SB_EVENT_TYPE_PASS:
pass_event_kwargs = _parse_pass(
pass_dict=raw_event["pass"],
team=team,
fidelity_version=fidelity_version,
)
event = PassEvent(
# TODO: Consider moving this to _parse_pass
receive_timestamp=timestamp + raw_event["duration"],
**pass_event_kwargs,
**generic_event_kwargs,
)
elif event_type == SB_EVENT_TYPE_SHOT:
shot_event_kwargs = _parse_shot(
shot_dict=raw_event["shot"]
)
event = ShotEvent(
**shot_event_kwargs, **generic_event_kwargs
)
# For dribble and carry the definitions
# are flipped between Statsbomb and kloppy
elif event_type == SB_EVENT_TYPE_DRIBBLE:
take_on_event_kwargs = _parse_take_on(
take_on_dict=raw_event["dribble"]
)
event = TakeOnEvent(
**take_on_event_kwargs, **generic_event_kwargs
)
elif event_type == SB_EVENT_TYPE_CARRY:
carry_event_kwargs = _parse_carry(
carry_dict=raw_event["carry"],
fidelity_version=fidelity_version,
)
event = CarryEvent(
# TODO: Consider moving this to _parse_carry
end_timestamp=timestamp + raw_event["duration"],
**carry_event_kwargs,
**generic_event_kwargs,
)
else:
event = GenericEvent(
result=None,
event_name=raw_event["type"]["name"],
**generic_event_kwargs,
)
if (
not wanted_event_types
or event.event_type in wanted_event_types
):
events.append(event)
metadata = Metadata(
teams=teams,
periods=periods,
pitch_dimensions=PitchDimensions(
x_dim=Dimension(0, 120), y_dim=Dimension(0, 80)
),
frame_rate=None,
orientation=Orientation.ACTION_EXECUTING_TEAM,
flags=DatasetFlag.BALL_OWNING_TEAM,
score=None,
)
return EventDataset(metadata=metadata, records=events,)
def _frame_from_line(cls, teams, period, line, frame_rate):
line = str(line)
frame_id, players, ball = line.strip().split(":")[:3]
players_data = {}
for player_data in players.split(";")[:-1]:
team_id, target_id, jersey_no, x, y, speed = player_data.split(",")
team_id = int(team_id)
if team_id == 1:
team = teams[0]
elif team_id == 0:
team = teams[1]
else:
# it's probably -1, but make sure it doesn't crash
continue
player = team.get_player_by_jersey_number(jersey_no)
if not player:
player = Player(
player_id=f"{team.ground}_{jersey_no}",
team=team,
jersey_no=int(jersey_no),
)
team.players.append(player)
players_data[player] = PlayerData(
coordinates=Point(float(x), float(y)), speed=float(speed)
)
(
ball_x,
ball_y,
ball_z,
ball_speed,
ball_owning_team,
ball_state,
) = ball.rstrip(";").split(",")[:6]
frame_id = int(frame_id)
if ball_owning_team == "H":
ball_owning_team = teams[0]
elif ball_owning_team == "A":
ball_owning_team = teams[1]
else:
raise DeserializationError(
f"Unknown ball owning team: {ball_owning_team}"
)
if ball_state == "Alive":
ball_state = BallState.ALIVE
elif ball_state == "Dead":
ball_state = BallState.DEAD
else:
raise DeserializationError(f"Unknown ball state: {ball_state}")
return Frame(
frame_id=frame_id,
timestamp=frame_id / frame_rate - period.start_timestamp,
ball_coordinates=Point3D(
float(ball_x), float(ball_y), float(ball_z)
),
ball_state=ball_state,
ball_owning_team=ball_owning_team,
players_data=players_data,
period=period,
other_data={},
)
def deserialize(self, inputs: DatafactoryInputs) -> EventDataset:
transformer = self.get_transformer(length=2, width=2)
with performance_logging("load data", logger=logger):
data = json.load(inputs.event_data)
match = data["match"]
score_data = data["scoreStatus"]
incidences = data["incidences"]
players_data = data["players"]
teams_data = data["teams"]
with performance_logging("parse data", logger=logger):
teams = []
scores = []
team_ids = (
(Ground.HOME, str(match["homeTeamId"])),
(Ground.AWAY, str(match["awayTeamId"])),
)
for ground, team_id in team_ids:
team = Team(
team_id=team_id,
name=teams_data[team_id]["name"],
ground=ground,
)
team.players = [
Player(
player_id=player_id,
team=team,
first_name=player["name"]["first"],
last_name=player["name"]["last"],
name=player["name"]["shortName"]
or player["name"]["nick"],
jersey_no=player["squadNo"],
starting=not player["substitute"],
) for player_id, player in players_data.items()
if str(player["teamId"]) == team_id
]
teams.append(team)
scores.append(score_data.get(team_id, {}).get("score"))
score = Score(home=scores[0], away=scores[1])
# setup periods
status = incidences.pop(DF_EVENT_CLASS_STATUS)
# start timestamps are fixed
start_ts = {1: 0, 2: 45 * 60, 3: 90 * 60, 4: 105 * 60, 5: 120 * 60}
# check for end status updates to setup periods
end_event_types = {
DF_EVENT_TYPE_STATUS_MATCH_END,
DF_EVENT_TYPE_STATUS_FIRST_HALF_END,
DF_EVENT_TYPE_STATUS_SECOND_HALF_END,
DF_EVENT_TYPE_STATUS_FIRST_EXTRA_END,
DF_EVENT_TYPE_STATUS_SECOND_EXTRA_END,
}
periods = {}
for status_update in status.values():
if status_update["type"] not in end_event_types:
continue
half = status_update["t"]["half"]
end_ts = parse_str_ts(status_update)
periods[half] = Period(
id=half,
start_timestamp=start_ts[half],
end_timestamp=end_ts,
attacking_direction=AttackingDirection.HOME_AWAY if half %
2 == 1 else AttackingDirection.AWAY_HOME,
)
# exclude goals, already listed as shots too
incidences.pop(DF_EVENT_CLASS_GOALS)
raw_events = [(k, e_id, e) for k in incidences
for e_id, e in incidences[k].items()]
# sort events by timestamp, event_id
raw_events.sort(key=lambda e: (
e[2]["t"]["half"],
e[2]["t"]["m"],
e[2]["t"]["s"] or 0,
e[1],
))
home_team, away_team = teams
events = []
previous_event = next_event = None
for i, (e_class, e_id, raw_event) in enumerate(raw_events):
period = periods.get(raw_event["t"]["half"])
if period is None:
# skip invalid event
continue
timestamp = parse_str_ts(raw_event)
if (previous_event is not None and
previous_event["t"]["half"] != raw_event["t"]["half"]):
previous_event = None
next_event = (raw_events[i + 1][2]
if i + 1 < len(raw_events) else None)
team, player = _get_team_and_player(raw_event, home_team,
away_team)
event_base_kwargs = dict(
# from DataRecord
period=period,
timestamp=timestamp,
ball_owning_team=team,
ball_state=BallState.ALIVE,
# from Event
event_id=e_id,
team=team,
player=player,
coordinates=(_parse_coordinates(raw_event["coord"]["1"])
if "coord" in raw_event else None),
raw_event=raw_event,
result=None,
qualifiers=None,
)
if e_class in DF_EVENT_CLASS_PASSES:
pass_event_kwargs = _parse_pass(
raw_event=raw_event,
team=team,
previous_event=previous_event,
next_event=next_event,
)
event_base_kwargs.update(pass_event_kwargs)
event = PassEvent.create(**event_base_kwargs)
elif e_class == DF_EVENT_CLASS_SHOTS:
shot_event_kwargs = _parse_shot(
raw_event=raw_event,
previous_event=previous_event,
)
event_base_kwargs.update(shot_event_kwargs)
event = ShotEvent.create(**event_base_kwargs)
elif e_class == DF_EVENT_CLASS_STEALINGS:
event = RecoveryEvent.create(**event_base_kwargs)
elif e_class == DF_EVENT_CLASS_FOULS:
# NOTE: could use qualifiers? (hand, foul, penalty?)
# switch possession team
event_base_kwargs["ball_owning_team"] = (
home_team if team == away_team else away_team)
event = FoulCommittedEvent.create(**event_base_kwargs)
elif e_class in DF_EVENT_CLASS_CARDS:
card_kwargs = _parse_card(raw_event=raw_event, )
event_base_kwargs.update(card_kwargs)
event = CardEvent.create(**event_base_kwargs)
elif e_class == DF_EVENT_CLASS_SUBSTITUTIONS:
substitution_event_kwargs = _parse_substitution(
raw_event=raw_event, team=team)
event_base_kwargs.update(substitution_event_kwargs)
event = SubstitutionEvent.create(**event_base_kwargs)
else:
# otherwise, a generic event
event = GenericEvent.create(
event_name=e_class,
**event_base_kwargs,
)
# check if the event implies ball was out of the field and add a synthetic out event
if raw_event["type"] in BALL_OUT_EVENTS:
ball_out_event = BallOutEvent.create(
# from DataRecord
period=period,
timestamp=timestamp,
ball_owning_team=team,
ball_state=BallState.DEAD,
# from Event
event_id=e_id,
team=team,
player=player,
coordinates=event.coordinates,
raw_event=raw_event,
result=None,
qualifiers=None,
)
if self.should_include_event(event):
events.append(
transformer.transform_event(ball_out_event))
if self.should_include_event(event):
events.append(transformer.transform_event(event))
# only consider as a previous_event a ball-in-play event
if e_class not in (
DF_EVENT_CLASS_YELLOW_CARDS,
DF_EVENT_CLASS_RED_CARDS,
DF_EVENT_CLASS_SUBSTITUTIONS,
DF_EVENT_CLASS_PENALTY_SHOOTOUT,
):
previous_event = raw_event
metadata = Metadata(
teams=teams,
periods=sorted(periods.values(), key=lambda p: p.id),
pitch_dimensions=transformer.get_to_coordinate_system().
pitch_dimensions,
frame_rate=None,
orientation=Orientation.HOME_TEAM,
flags=DatasetFlag.BALL_OWNING_TEAM,
score=score,
provider=Provider.DATAFACTORY,
coordinate_system=transformer.get_to_coordinate_system(),
)
return EventDataset(
metadata=metadata,
records=events,
)
def deserialize(self, inputs: SkillCornerInputs) -> TrackingDataset:
metadata = self.__load_json(inputs.meta_data)
raw_data = self.__load_json(inputs.raw_data)
with performance_logging("Loading metadata", logger=logger):
periods = self.__get_periods(raw_data)
teamdict = {
metadata["home_team"].get("id"): "home_team",
metadata["away_team"].get("id"): "away_team",
}
player_id_to_team_dict = {
player["trackable_object"]: player["team_id"]
for player in metadata["players"]
}
player_dict = {
player["trackable_object"]: player
for player in metadata["players"]
}
referee_dict = {
ref["trackable_object"]: "referee"
for ref in metadata["referees"]
}
ball_id = metadata["ball"]["trackable_object"]
# there are different pitch_sizes in SkillCorner
pitch_size_width = metadata["pitch_width"]
pitch_size_length = metadata["pitch_length"]
transformer = self.get_transformer(length=pitch_size_length,
width=pitch_size_width)
home_team_id = metadata["home_team"]["id"]
away_team_id = metadata["away_team"]["id"]
players = {"HOME": {}, "AWAY": {}}
home_team = Team(
team_id=home_team_id,
name=metadata["home_team"]["name"],
ground=Ground.HOME,
)
away_team = Team(
team_id=away_team_id,
name=metadata["away_team"]["name"],
ground=Ground.AWAY,
)
teams = [home_team, away_team]
for player_id in player_dict.keys():
player = player_dict.get(player_id)
team_id = player["team_id"]
if team_id == home_team_id:
team_string = "HOME"
team = home_team
elif team_id == away_team_id:
team_string = "AWAY"
team = away_team
players[team_string][player_id] = Player(
player_id=f"{team.ground}_{player['number']}",
team=team,
jersey_no=player["number"],
name=f"{player['first_name']} {player['last_name']}",
first_name=player["first_name"],
last_name=player["last_name"],
starting=player["start_time"] == "00:00:00",
position=Position(
position_id=player["player_role"].get("id"),
name=player["player_role"].get("name"),
coordinates=None,
),
attributes={},
)
home_team.players = list(players["HOME"].values())
away_team.players = list(players["AWAY"].values())
anon_players = {"HOME": {}, "AWAY": {}}
with performance_logging("Loading data", logger=logger):
def _iter():
n = 0
sample = 1.0 / self.sample_rate
for frame in raw_data:
frame_period = frame["period"]
if frame_period is not None:
if n % sample == 0:
yield frame
n += 1
frames = []
n_frames = 0
for _frame in _iter():
# include frame if there is any tracking data, players or ball.
# or if include_empty_frames == True
if self.include_empty_frames or len(_frame["data"]) > 0:
frame = self._get_frame_data(
teams,
teamdict,
players,
player_id_to_team_dict,
periods,
player_dict,
anon_players,
ball_id,
referee_dict,
_frame,
)
frame = transformer.transform_frame(frame)
frames.append(frame)
n_frames += 1
if self.limit and n_frames >= self.limit:
break
self._set_skillcorner_attacking_directions(frames, periods)
frame_rate = 10
orientation = (Orientation.HOME_TEAM if periods[1].attacking_direction
== AttackingDirection.HOME_AWAY else
Orientation.AWAY_TEAM)
metadata = Metadata(
teams=teams,
periods=periods,
pitch_dimensions=transformer.get_to_coordinate_system().
pitch_dimensions,
score=Score(
home=metadata["home_team_score"],
away=metadata["away_team_score"],
),
frame_rate=frame_rate,
orientation=orientation,
provider=Provider.SKILLCORNER,
flags=~(DatasetFlag.BALL_STATE | DatasetFlag.BALL_OWNING_TEAM),
coordinate_system=transformer.get_to_coordinate_system(),
)
return TrackingDataset(
records=frames,
metadata=metadata,
)
def __create_iterator(
self,
data: Readable,
sample_rate: float,
frame_rate: int,
ground: Ground,
) -> Iterator:
"""
Notes:
1. the y-axis is flipped because Metrica use (y, -y) instead of (-y, y)
"""
team = None
frame_idx = 0
frame_sample = 1 / sample_rate
player_jersey_numbers = []
period = None
for i, line in enumerate(data):
line = line.strip().decode("ascii")
columns = line.split(",")
if i == 0:
team_name = columns[3]
team = Team(team_id=str(ground), name=team_name, ground=ground)
elif i == 1:
player_jersey_numbers = columns[3:-2:2]
players = [
Player(
player_id=f"{team.ground}_{jersey_number}",
jersey_no=int(jersey_number),
team=team,
)
for jersey_number in player_jersey_numbers
]
team.players = players
elif i == 2:
# consider doing some validation on the columns
pass
else:
period_id = int(columns[0])
frame_id = int(columns[1])
if period is None or period.id != period_id:
period = Period(
id=period_id,
start_timestamp=frame_id / frame_rate,
end_timestamp=frame_id / frame_rate,
)
else:
# consider not update this every frame for performance reasons
period.end_timestamp = frame_id / frame_rate
if frame_idx % frame_sample == 0:
yield self.__PartialFrame(
team=team,
period=period,
frame_id=frame_id,
players_coordinates={
player: Point(
x=float(columns[3 + i * 2]),
y=1 - float(columns[3 + i * 2 + 1]),
)
for i, player in enumerate(players)
if columns[3 + i * 2] != "NaN"
},
ball_coordinates=Point(
x=float(columns[-2]), y=1 - float(columns[-1])
)
if columns[-2] != "NaN"
else None,
)
frame_idx += 1
def deserialize(self, inputs: SecondSpectrumInputs) -> TrackingDataset:
metadata = None
# Handles the XML metadata that contains the pitch dimensions and frame info
with performance_logging("Loading XML metadata", logger=logger):
# The meta data can also be in JSON format. In that case
# it also contains the 'additional metadata'.
# First do a 'peek' to determine the char
first_byte = inputs.meta_data.read(1)
if first_byte == b"{":
metadata = json.loads(first_byte + inputs.meta_data.read())
frame_rate = int(metadata["fps"])
pitch_size_height = float(metadata["pitchLength"])
pitch_size_width = float(metadata["pitchWidth"])
periods = []
for period in metadata["periods"]:
start_frame_id = int(period["startFrameIdx"])
end_frame_id = int(period["endFrameIdx"])
if start_frame_id != 0 or end_frame_id != 0:
# Frame IDs are unix timestamps (in milliseconds)
periods.append(
Period(
id=int(period["number"]),
start_timestamp=start_frame_id,
end_timestamp=end_frame_id,
)
)
else:
match = objectify.fromstring(
first_byte + inputs.meta_data.read()
).match
frame_rate = int(match.attrib["iFrameRateFps"])
pitch_size_height = float(match.attrib["fPitchYSizeMeters"])
pitch_size_width = float(match.attrib["fPitchXSizeMeters"])
periods = []
for period in match.iterchildren(tag="period"):
start_frame_id = int(period.attrib["iStartFrame"])
end_frame_id = int(period.attrib["iEndFrame"])
if start_frame_id != 0 or end_frame_id != 0:
# Frame IDs are unix timestamps (in milliseconds)
periods.append(
Period(
id=int(period.attrib["iId"]),
start_timestamp=start_frame_id,
end_timestamp=end_frame_id,
)
)
# Default team initialisation
home_team = Team(team_id="home", name="home", ground=Ground.HOME)
away_team = Team(team_id="away", name="away", ground=Ground.AWAY)
teams = [home_team, away_team]
if inputs.additional_meta_data or metadata:
with performance_logging("Loading JSON metadata", logger=logger):
try:
if inputs.additional_meta_data:
metadata = json.loads(
inputs.additional_meta_data.read()
)
home_team_id = metadata["homeOptaId"]
away_team_id = metadata["awayOptaId"]
# Tries to parse (short) team names from the description string
try:
home_name = (
metadata["description"].split("-")[0].strip()
)
away_name = (
metadata["description"]
.split("-")[1]
.split(":")[0]
.strip()
)
except:
home_name, away_name = "home", "away"
teams[0].team_id = home_team_id
teams[0].name = home_name
teams[1].team_id = away_team_id
teams[1].name = away_name
for team, team_str in zip(
teams, ["homePlayers", "awayPlayers"]
):
for player_data in metadata[team_str]:
# We use the attributes field of Player to store the extra IDs provided by the
# metadata. We designate the player_id to be the 'optaId' field as this is what's
# used as 'player_id' in the raw frame data file
player_attributes = {
k: v
for k, v in player_data.items()
if k in ["ssiId", "optaUuid"]
}
player = Player(
player_id=player_data["optaId"],
name=player_data["name"],
starting=player_data["position"] != "SUB",
position=player_data["position"],
team=team,
jersey_no=int(player_data["number"]),
attributes=player_attributes,
)
team.players.append(player)
except: # TODO: More specific exception
logging.warning(
"Optional JSON Metadata is malformed. Continuing without"
)
# Handles the tracking frame data
with performance_logging("Loading data", logger=logger):
transformer = self.get_transformer(
length=pitch_size_width, width=pitch_size_height
)
def _iter():
n = 0
sample = 1 / self.sample_rate
for line_ in inputs.raw_data.readlines():
line_ = line_.strip().decode("ascii")
if not line_:
continue
# Each line is just json so we just parse it
frame_data = json.loads(line_)
if self.only_alive and not frame_data["live"]:
continue
if n % sample == 0:
yield frame_data
n += 1
frames = []
for n, frame_data in enumerate(_iter()):
period = periods[frame_data["period"] - 1]
frame = self._frame_from_framedata(teams, period, frame_data)
frame = transformer.transform_frame(frame)
frames.append(frame)
if not period.attacking_direction_set:
period.set_attacking_direction(
attacking_direction=attacking_direction_from_frame(
frame
)
)
if self.limit and n + 1 >= self.limit:
break
orientation = (
Orientation.FIXED_HOME_AWAY
if periods[0].attacking_direction == AttackingDirection.HOME_AWAY
else Orientation.FIXED_AWAY_HOME
)
metadata = Metadata(
teams=teams,
periods=periods,
pitch_dimensions=transformer.get_to_coordinate_system().pitch_dimensions,
score=None,
frame_rate=frame_rate,
orientation=orientation,
provider=Provider.SECONDSPECTRUM,
flags=DatasetFlag.BALL_OWNING_TEAM | DatasetFlag.BALL_STATE,
coordinate_system=transformer.get_to_coordinate_system(),
)
return TrackingDataset(
records=frames,
metadata=metadata,
)
def deserialize(self, inputs: StatsbombInputs) -> EventDataset:
transformer = self.get_transformer(length=120, width=80)
with performance_logging("load data", logger=logger):
raw_events = json.load(inputs.event_data)
home_lineup, away_lineup = json.load(inputs.lineup_data)
(
shot_fidelity_version,
xy_fidelity_version,
) = _determine_xy_fidelity_versions(raw_events)
logger.info(
f"Determined Fidelity versions: shot v{shot_fidelity_version} / XY v{xy_fidelity_version}"
)
with performance_logging("parse data", logger=logger):
starting_player_ids = {
str(player["player"]["id"])
for raw_event in raw_events
if raw_event["type"]["id"] == SB_EVENT_TYPE_STARTING_XI
for player in raw_event["tactics"]["lineup"]
}
starting_formations = {
raw_event["team"]["id"]: FormationType("-".join(
list(str(raw_event["tactics"]["formation"]))))
for raw_event in raw_events
if raw_event["type"]["id"] == SB_EVENT_TYPE_STARTING_XI
}
home_team = Team(
team_id=str(home_lineup["team_id"]),
name=home_lineup["team_name"],
ground=Ground.HOME,
starting_formation=starting_formations[home_lineup["team_id"]],
)
home_team.players = [
Player(
player_id=str(player["player_id"]),
team=home_team,
name=player["player_name"],
jersey_no=int(player["jersey_number"]),
starting=str(player["player_id"]) in starting_player_ids,
) for player in home_lineup["lineup"]
]
away_team = Team(
team_id=str(away_lineup["team_id"]),
name=away_lineup["team_name"],
ground=Ground.AWAY,
starting_formation=starting_formations[away_lineup["team_id"]],
)
away_team.players = [
Player(
player_id=str(player["player_id"]),
team=away_team,
name=player["player_name"],
jersey_no=int(player["jersey_number"]),
starting=str(player["player_id"]) in starting_player_ids,
) for player in away_lineup["lineup"]
]
teams = [home_team, away_team]
periods = []
period = None
events = []
for raw_event in raw_events:
if raw_event["team"]["id"] == home_lineup["team_id"]:
team = home_team
elif raw_event["team"]["id"] == away_lineup["team_id"]:
team = away_team
else:
raise DeserializationError(
f"Unknown team_id {raw_event['team']['id']}")
if (raw_event["possession_team"]["id"] ==
home_lineup["team_id"]):
possession_team = home_team
elif (raw_event["possession_team"]["id"] ==
away_lineup["team_id"]):
possession_team = away_team
else:
raise DeserializationError(
f"Unknown possession_team_id: {raw_event['possession_team']}"
)
timestamp = parse_str_ts(raw_event["timestamp"])
period_id = int(raw_event["period"])
if not period or period.id != period_id:
period = Period(
id=period_id,
start_timestamp=(
timestamp if not period
# period = [start, end], add millisecond to prevent overlapping
else timestamp + period.end_timestamp + 0.001),
end_timestamp=None,
)
periods.append(period)
else:
period.end_timestamp = period.start_timestamp + timestamp
player = None
if "player" in raw_event:
player = team.get_player_by_id(raw_event["player"]["id"])
event_type = raw_event["type"]["id"]
if event_type == SB_EVENT_TYPE_SHOT:
fidelity_version = shot_fidelity_version
elif event_type in (
SB_EVENT_TYPE_CARRY,
SB_EVENT_TYPE_DRIBBLE,
SB_EVENT_TYPE_PASS,
):
fidelity_version = xy_fidelity_version
else:
# TODO: Uh ohhhh.. don't know which one to pick
fidelity_version = xy_fidelity_version
generic_event_kwargs = {
# from DataRecord
"period":
period,
"timestamp":
timestamp,
"ball_owning_team":
possession_team,
"ball_state":
BallState.ALIVE,
# from Event
"event_id":
raw_event["id"],
"team":
team,
"player":
player,
"coordinates": (_parse_coordinates(
raw_event.get("location"),
fidelity_version,
) if "location" in raw_event else None),
"related_event_ids":
raw_event.get("related_events", []),
"raw_event":
raw_event,
}
new_events = []
if event_type == SB_EVENT_TYPE_PASS:
pass_event_kwargs = _parse_pass(
pass_dict=raw_event["pass"],
team=team,
fidelity_version=fidelity_version,
)
pass_event = PassEvent.create(
# TODO: Consider moving this to _parse_pass
receive_timestamp=timestamp + raw_event["duration"],
**pass_event_kwargs,
**generic_event_kwargs,
)
new_events.append(pass_event)
elif event_type == SB_EVENT_TYPE_SHOT:
shot_event_kwargs = _parse_shot(
shot_dict=raw_event["shot"], )
shot_event = ShotEvent.create(
**shot_event_kwargs,
**generic_event_kwargs,
)
new_events.append(shot_event)
# For dribble and carry the definitions
# are flipped between Statsbomb and kloppy
elif event_type == SB_EVENT_TYPE_DRIBBLE:
take_on_event_kwargs = _parse_take_on(
take_on_dict=raw_event["dribble"], )
take_on_event = TakeOnEvent.create(
qualifiers=None,
**take_on_event_kwargs,
**generic_event_kwargs,
)
new_events.append(take_on_event)
elif event_type == SB_EVENT_TYPE_CARRY:
carry_event_kwargs = _parse_carry(
carry_dict=raw_event["carry"],
fidelity_version=fidelity_version,
)
carry_event = CarryEvent.create(
qualifiers=None,
# TODO: Consider moving this to _parse_carry
end_timestamp=timestamp + raw_event.get("duration", 0),
**carry_event_kwargs,
**generic_event_kwargs,
)
new_events.append(carry_event)
# lineup affecting events
elif event_type == SB_EVENT_TYPE_SUBSTITUTION:
substitution_event_kwargs = _parse_substitution(
substitution_dict=raw_event["substitution"],
team=team,
)
substitution_event = SubstitutionEvent.create(
result=None,
qualifiers=None,
**substitution_event_kwargs,
**generic_event_kwargs,
)
new_events.append(substitution_event)
elif event_type == SB_EVENT_TYPE_BAD_BEHAVIOUR:
bad_behaviour_kwargs = _parse_bad_behaviour(
bad_behaviour_dict=raw_event.get("bad_behaviour",
{}), )
if "card" in bad_behaviour_kwargs:
card_kwargs = bad_behaviour_kwargs["card"]
card_event = CardEvent.create(
result=None,
qualifiers=None,
card_type=card_kwargs["card_type"],
**generic_event_kwargs,
)
new_events.append(card_event)
elif event_type == SB_EVENT_TYPE_FOUL_COMMITTED:
foul_committed_kwargs = _parse_foul_committed(
foul_committed_dict=raw_event.get(
"foul_committed", {}), )
foul_committed_event = FoulCommittedEvent.create(
result=None,
qualifiers=None,
**generic_event_kwargs,
)
new_events.append(foul_committed_event)
if "card" in foul_committed_kwargs:
card_kwargs = foul_committed_kwargs["card"]
card_event = CardEvent.create(
result=None,
qualifiers=None,
card_type=card_kwargs["card_type"],
**generic_event_kwargs,
)
new_events.append(card_event)
elif event_type == SB_EVENT_TYPE_PLAYER_ON:
player_on_event = PlayerOnEvent.create(
result=None,
qualifiers=None,
**generic_event_kwargs,
)
new_events.append(player_on_event)
elif event_type == SB_EVENT_TYPE_PLAYER_OFF:
player_off_event = PlayerOffEvent.create(
result=None,
qualifiers=None,
**generic_event_kwargs,
)
new_events.append(player_off_event)
elif event_type == SB_EVENT_TYPE_RECOVERY:
recovery_event = RecoveryEvent.create(
result=None,
qualifiers=None,
**generic_event_kwargs,
)
new_events.append(recovery_event)
elif event_type == SB_EVENT_TYPE_FORMATION_CHANGE:
formation_change_event_kwargs = _parse_formation_change(
raw_event["tactics"]["formation"])
formation_change_event = FormationChangeEvent.create(
result=None,
qualifiers=None,
**formation_change_event_kwargs,
**generic_event_kwargs,
)
new_events.append(formation_change_event)
# rest: generic
else:
generic_event = GenericEvent.create(
result=None,
qualifiers=None,
event_name=raw_event["type"]["name"],
**generic_event_kwargs,
)
new_events.append(generic_event)
for event in new_events:
if self.should_include_event(event):
transformed_event = transformer.transform_event(event)
events.append(transformed_event)
# Checks if the event ended out of the field and adds a synthetic out event
if event.result in OUT_EVENT_RESULTS:
generic_event_kwargs["ball_state"] = BallState.DEAD
if event.receiver_coordinates:
generic_event_kwargs[
"coordinates"] = event.receiver_coordinates
ball_out_event = BallOutEvent.create(
result=None,
qualifiers=None,
**generic_event_kwargs,
)
if self.should_include_event(ball_out_event):
transformed_ball_out_event = (
transformer.transform_event(ball_out_event)
)
events.append(transformed_ball_out_event)
metadata = Metadata(
teams=teams,
periods=periods,
pitch_dimensions=transformer.get_to_coordinate_system().
pitch_dimensions,
frame_rate=None,
orientation=Orientation.ACTION_EXECUTING_TEAM,
flags=DatasetFlag.BALL_OWNING_TEAM,
score=None,
provider=Provider.STATSBOMB,
coordinate_system=transformer.get_to_coordinate_system(),
)
return EventDataset(
metadata=metadata,
records=events,
)
def _frame_from_line(cls, teams, period, line, frame_rate):
line = str(line)
frame_id, players, ball = line.strip().split(":")[:3]
players_coordinates = {}
for player_data in players.split(";")[:-1]:
team_id, target_id, jersey_no, x, y, speed = player_data.split(",")
team_id = int(team_id)
if team_id == 1:
team = teams[0]
elif team_id == 0:
team = teams[1]
else:
raise Exception(f"Unknown team {team_id}")
player = team.get_player_by_jersey_number(jersey_no)
if not player:
player = Player(
player_id=f"{team.ground}_{jersey_no}",
team=team,
jersey_no=int(jersey_no),
)
team.players.append(player)
players_coordinates[player] = Point(float(x), float(y))
(
ball_x,
ball_y,
ball_z,
ball_speed,
ball_owning_team,
ball_state,
) = ball.rstrip(";").split(",")[:6]
frame_id = int(frame_id)
if ball_owning_team == "H":
ball_owning_team = teams[0]
elif ball_owning_team == "A":
ball_owning_team = teams[1]
else:
raise Exception(f"Unknown ball owning team: {ball_owning_team}")
if ball_state == "Alive":
ball_state = BallState.ALIVE
elif ball_state == "Dead":
ball_state = BallState.DEAD
else:
raise Exception(f"Unknown ball state: {ball_state}")
return Frame(
frame_id=frame_id,
timestamp=frame_id / frame_rate - period.start_timestamp,
ball_coordinates=Point(float(ball_x), float(ball_y)),
ball_state=ball_state,
ball_owning_team=ball_owning_team,
players_coordinates=players_coordinates,
period=period,
)
def deserialize(self,
inputs: Dict[str, Readable],
options: Dict = None) -> TrackingDataset:
"""
Deserialize SkillCorner tracking data into a `TrackingDataset`.
Parameters
----------
inputs : dict
input `raw_data` should point to a `Readable` object containing
the 'json' formatted raw data. input `metadata` should point to
the json metadata data.
options : dict
Options for deserialization of the TRACAB file. Possible options are:
`include_empty_frames` (boolean): default = False to specify whether frames without
any players_coordinates or the ball_coordinates should be loaded
`sample_rate` (float between 0 and 1) to specify the amount of frames that should be loaded
and `limit` (int) to specify the max number of frames that will be returned.
Returns
-------
dataset : TrackingDataset
Raises
------
-
See Also
--------
Examples
--------
>>> serializer = SkillCornerSerializer()
>>> with open("match_data.json", "rb") as meta, \
>>> open("structured_data.json", "rb") as raw:
>>> dataset = serializer.deserialize(
>>> inputs={
>>> 'metadata': meta,
>>> 'raw_data': raw
>>> },
>>> options={
>>> }
>>> )
"""
self.__validate_inputs(inputs)
metadata = self.__load_json(inputs["metadata"])
raw_data = self.__load_json(inputs["raw_data"])
if not options:
options = {}
sample_rate = float(options.get("sample_rate", 1.0))
limit = int(options.get("limit", 0))
include_empty_frames = bool(options.get("include_empty_frames", False))
with performance_logging("Loading metadata", logger=logger):
periods = self.__get_periods(raw_data)
teamdict = {
metadata["home_team"].get("id"): "home_team",
metadata["away_team"].get("id"): "away_team",
}
player_id_to_team_dict = {
player["trackable_object"]: player["team_id"]
for player in metadata["players"]
}
player_dict = {
player["trackable_object"]: player
for player in metadata["players"]
}
referee_dict = {
ref["trackable_object"]: "referee"
for ref in metadata["referees"]
}
ball_id = metadata["ball"]["trackable_object"]
# there are different pitch_sizes in SkillCorner
pitch_size_width = metadata["pitch_width"]
pitch_size_length = metadata["pitch_length"]
home_team_id = metadata["home_team"]["id"]
away_team_id = metadata["away_team"]["id"]
players = {"HOME": {}, "AWAY": {}}
home_team = Team(
team_id=home_team_id,
name=metadata["home_team"]["name"],
ground=Ground.HOME,
)
self.home_team = home_team
away_team = Team(
team_id=away_team_id,
name=metadata["away_team"]["name"],
ground=Ground.AWAY,
)
self.away_team = away_team
teams = [home_team, away_team]
for player_id in player_dict.keys():
player = player_dict.get(player_id)
team_id = player["team_id"]
if team_id == home_team_id:
team_string = "HOME"
team = home_team
elif team_id == away_team_id:
team_string = "AWAY"
team = away_team
players[team_string][player_id] = Player(
player_id=f"{team.ground}_{player['number']}",
team=team,
jersey_no=player["number"],
name=f"{player['first_name']} {player['last_name']}",
first_name=player["first_name"],
last_name=player["last_name"],
starting=player["start_time"] == "00:00:00",
position=Position(
position_id=player["player_role"].get("id"),
name=player["player_role"].get("name"),
coordinates=None,
),
attributes={},
)
home_team.players = list(players["HOME"].values())
away_team.players = list(players["AWAY"].values())
anon_players = {"HOME": {}, "AWAY": {}}
with performance_logging("Loading data", logger=logger):
def _iter():
n = 0
sample = 1.0 / sample_rate
for frame in raw_data:
frame_period = frame["period"]
if frame_period is not None:
if n % sample == 0:
yield frame
n += 1
frames = []
n_frames = 0
for _frame in _iter():
# include frame if there is any tracking data, players or ball.
# or if include_empty_frames == True
if include_empty_frames or len(_frame["data"]) > 0:
frame = self._get_frame_data(
teams,
teamdict,
players,
player_id_to_team_dict,
periods,
player_dict,
anon_players,
ball_id,
referee_dict,
_frame,
)
frames.append(frame)
n_frames += 1
if limit and n_frames >= limit:
break
self._set_skillcorner_attacking_directions(frames, periods)
frame_rate = 10
orientation = (Orientation.HOME_TEAM if periods[1].attacking_direction
== AttackingDirection.HOME_AWAY else
Orientation.AWAY_TEAM)
metadata = Metadata(
teams=teams,
periods=periods,
pitch_dimensions=PitchDimensions(
x_dim=Dimension(-(pitch_size_length / 2),
(pitch_size_length / 2)),
y_dim=Dimension(-(pitch_size_width / 2),
(pitch_size_width / 2)),
x_per_meter=1,
y_per_meter=1,
),
score=Score(
home=metadata["home_team_score"],
away=metadata["away_team_score"],
),
frame_rate=frame_rate,
orientation=orientation,
provider=Provider.SKILLCORNER,
flags=~(DatasetFlag.BALL_STATE | DatasetFlag.BALL_OWNING_TEAM),
)
return TrackingDataset(
records=frames,
metadata=metadata,
)