def deserialize(self, inputs: SportsCodeInputs) -> CodeDataset:
all_instances = objectify.fromstring(inputs.data.read())
codes = []
period = Period(id=1, start_timestamp=0, end_timestamp=0)
for instance in all_instances.ALL_INSTANCES.iterchildren():
end_timestamp = float(instance.end)
code = Code(
period=period,
code_id=str(instance.ID),
code=str(instance.code),
timestamp=float(instance.start),
end_timestamp=end_timestamp,
labels={
str(label.find("group")):
parse_value(str(label.find("text")))
for label in instance.iterchildren("label")
},
ball_state=None,
ball_owning_team=None,
)
period.end_timestamp = end_timestamp
codes.append(code)
return CodeDataset(
metadata=Metadata(
teams=[],
periods=[period],
pitch_dimensions=None,
score=Score(0, 0),
frame_rate=0.0,
orientation=Orientation.NOT_SET,
flags=~(DatasetFlag.BALL_OWNING_TEAM | DatasetFlag.BALL_STATE),
provider=Provider.OTHER,
coordinate_system=None,
),
records=codes,
)
def deserialize(self,
inputs: Dict[str, Readable],
options: Dict = None) -> EventDataset:
"""
Deserialize Opta event data into a `EventDataset`.
Parameters
----------
inputs : dict
input `f24_data` should point to a `Readable` object containing
the 'xml' formatted event data. input `f7_data` should point
to a `Readable` object containing the 'xml' formatted f7 data.
options : dict
Options for deserialization of the Opta 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 = OptaSerializer()
>>> with open("123_f24.xml", "rb") as f24_data, \
>>> open("123_f7.xml", "rb") as f7_data:
>>>
>>> dataset = serializer.deserialize(
>>> inputs={
>>> 'f24_data': f24_data,
>>> 'f7_data': f7_data
>>> },
>>> options={
>>> 'event_types': ["pass", "take_on", "carry", "shot"]
>>> }
>>> )
"""
self.__validate_inputs(inputs)
if not options:
options = {}
with performance_logging("load data", logger=logger):
f7_root = objectify.fromstring(inputs["f7_data"].read())
f24_root = objectify.fromstring(inputs["f24_data"].read())
wanted_event_types = [
EventType[event_type.upper()]
for event_type in options.get("event_types", [])
]
with performance_logging("parse data", logger=logger):
matchdata_path = objectify.ObjectPath(
"SoccerFeed.SoccerDocument.MatchData")
team_elms = list(
matchdata_path.find(f7_root).iterchildren("TeamData"))
home_score = None
away_score = None
for team_elm in team_elms:
if team_elm.attrib["Side"] == "Home":
home_score = team_elm.attrib["Score"]
home_team = _team_from_xml_elm(team_elm, f7_root)
elif team_elm.attrib["Side"] == "Away":
away_score = team_elm.attrib["Score"]
away_team = _team_from_xml_elm(team_elm, f7_root)
else:
raise Exception(f"Unknown side: {team_elm.attrib['Side']}")
score = Score(home=home_score, away=away_score)
teams = [home_team, away_team]
if len(home_team.players) == 0 or len(away_team.players) == 0:
raise Exception("LineUp incomplete")
game_elm = f24_root.find("Game")
periods = [
Period(
id=1,
start_timestamp=None,
end_timestamp=None,
),
Period(
id=2,
start_timestamp=None,
end_timestamp=None,
),
]
possession_team = None
events = []
for event_elm in game_elm.iterchildren("Event"):
event_id = event_elm.attrib["id"]
type_id = int(event_elm.attrib["type_id"])
timestamp = _parse_f24_datetime(event_elm.attrib["timestamp"])
period_id = int(event_elm.attrib["period_id"])
for period in periods:
if period.id == period_id:
break
else:
logger.debug(
f"Skipping event {event_id} because period doesn't match {period_id}"
)
continue
if type_id == EVENT_TYPE_START_PERIOD:
logger.debug(
f"Set start of period {period.id} to {timestamp}")
period.start_timestamp = timestamp
elif type_id == EVENT_TYPE_END_PERIOD:
logger.debug(
f"Set end of period {period.id} to {timestamp}")
period.end_timestamp = timestamp
else:
if not period.start_timestamp:
# not started yet
continue
if event_elm.attrib["team_id"] == home_team.team_id:
team = teams[0]
elif event_elm.attrib["team_id"] == away_team.team_id:
team = teams[1]
else:
raise Exception(
f"Unknown team_id {event_elm.attrib['team_id']}")
x = float(event_elm.attrib["x"])
y = float(event_elm.attrib["y"])
outcome = int(event_elm.attrib["outcome"])
qualifiers = {
int(qualifier_elm.attrib["qualifier_id"]):
qualifier_elm.attrib.get("value")
for qualifier_elm in event_elm.iterchildren("Q")
}
player = None
if "player_id" in event_elm.attrib:
player = team.get_player_by_id(
event_elm.attrib["player_id"])
if type_id in BALL_OWNING_EVENTS:
possession_team = team
generic_event_kwargs = dict(
# from DataRecord
period=period,
timestamp=timestamp - period.start_timestamp,
ball_owning_team=possession_team,
ball_state=BallState.ALIVE,
# from Event
event_id=event_id,
team=team,
player=player,
coordinates=Point(x=x, y=y),
raw_event=event_elm,
)
if type_id == EVENT_TYPE_PASS:
pass_event_kwargs = _parse_pass(qualifiers, outcome)
event = PassEvent(
**pass_event_kwargs,
**generic_event_kwargs,
)
elif type_id == EVENT_TYPE_OFFSIDE_PASS:
pass_event_kwargs = _parse_offside_pass()
event = PassEvent(
**pass_event_kwargs,
**generic_event_kwargs,
)
elif type_id == EVENT_TYPE_TAKE_ON:
take_on_event_kwargs = _parse_take_on(outcome)
event = TakeOnEvent(
**take_on_event_kwargs,
**generic_event_kwargs,
)
elif type_id in (
EVENT_TYPE_SHOT_MISS,
EVENT_TYPE_SHOT_POST,
EVENT_TYPE_SHOT_SAVED,
EVENT_TYPE_SHOT_GOAL,
):
shot_event_kwargs = _parse_shot(
qualifiers,
type_id,
coordinates=generic_event_kwargs["coordinates"],
)
kwargs = {}
kwargs.update(generic_event_kwargs)
kwargs.update(shot_event_kwargs)
event = ShotEvent(**kwargs)
else:
event = GenericEvent(
**generic_event_kwargs,
result=None,
event_name=_get_event_type_name(type_id),
)
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, 100),
y_dim=Dimension(0, 100)),
score=score,
frame_rate=None,
orientation=Orientation.ACTION_EXECUTING_TEAM,
flags=DatasetFlag.BALL_OWNING_TEAM,
provider=Provider.OPTA,
)
return EventDataset(
metadata=metadata,
records=events,
)
def load_metadata(metadata_file: Readable,
provider: Provider = None) -> EPTSMetadata:
root = objectify.fromstring(metadata_file.read())
metadata = root.find("Metadata")
score_path = objectify.ObjectPath(
"Metadata.Sessions.Session[0].MatchParameters.Score")
score_elm = score_path.find(metadata)
score = Score(home=score_elm.LocalTeamScore,
away=score_elm.VisitingTeamScore)
_team_map = {
Ground.HOME: score_elm.attrib["idLocalTeam"],
Ground.AWAY: score_elm.attrib["idVisitingTeam"],
}
_team_name_map = {
team_elm.attrib["id"]: str(team_elm.find("Name"))
for team_elm in metadata.find("Teams").iterchildren(tag="Team")
}
teams_metadata = {}
for ground, team_id in _team_map.items():
team = Team(team_id=team_id,
name=_team_name_map[team_id],
ground=ground)
team.players = _load_players(metadata.find("Players"), team)
teams_metadata.update({ground: team})
data_format_specifications = _load_data_format_specifications(
root.find("DataFormatSpecifications"))
device_path = objectify.ObjectPath("Metadata.Devices.Device[0].Sensors")
sensors = _load_sensors(device_path.find(metadata))
_channel_map = {
channel.channel_id: channel
for sensor in sensors for channel in sensor.channels
}
_all_players = [
player for key, value in teams_metadata.items()
for player in value.players
]
_player_map = {player.player_id: player for player in _all_players}
player_channels = [
PlayerChannel(
player_channel_id=player_channel_elm.attrib["id"],
player=_player_map[player_channel_elm.attrib["playerId"]],
channel=_channel_map[player_channel_elm.attrib["channelId"]],
)
for player_channel_elm in metadata.find("PlayerChannels").iterchildren(
tag="PlayerChannel")
]
frame_rate = int(metadata.find("GlobalConfig").find("FrameRate"))
pitch_dimensions = _load_pitch_dimensions(metadata, sensors)
periods = _load_periods(metadata.find("GlobalConfig"), frame_rate)
if periods:
start_attacking_direction = periods[0].attacking_direction
else:
start_attacking_direction = None
orientation = (
(Orientation.FIXED_HOME_AWAY if start_attacking_direction
== AttackingDirection.HOME_AWAY else Orientation.FIXED_AWAY_HOME)
if start_attacking_direction != AttackingDirection.NOT_SET else None)
metadata.orientation = orientation
return EPTSMetadata(
teams=list(teams_metadata.values()),
periods=periods,
pitch_dimensions=pitch_dimensions,
data_format_specifications=data_format_specifications,
player_channels=player_channels,
frame_rate=frame_rate,
sensors=sensors,
score=score,
orientation=None,
provider=provider,
flags=~(DatasetFlag.BALL_STATE | DatasetFlag.BALL_OWNING_TEAM),
)
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 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,
)
def deserialize(self, inputs: OptaInputs) -> EventDataset:
transformer = self.get_transformer(length=100, width=100)
with performance_logging("load data", logger=logger):
f7_root = objectify.fromstring(inputs.f7_data.read())
f24_root = objectify.fromstring(inputs.f24_data.read())
with performance_logging("parse data", logger=logger):
matchdata_path = objectify.ObjectPath(
"SoccerFeed.SoccerDocument.MatchData")
team_elms = list(
matchdata_path.find(f7_root).iterchildren("TeamData"))
home_score = None
away_score = None
for team_elm in team_elms:
if team_elm.attrib["Side"] == "Home":
home_score = team_elm.attrib["Score"]
home_team = _team_from_xml_elm(team_elm, f7_root)
elif team_elm.attrib["Side"] == "Away":
away_score = team_elm.attrib["Score"]
away_team = _team_from_xml_elm(team_elm, f7_root)
else:
raise DeserializationError(
f"Unknown side: {team_elm.attrib['Side']}")
score = Score(home=home_score, away=away_score)
teams = [home_team, away_team]
if len(home_team.players) == 0 or len(away_team.players) == 0:
raise DeserializationError("LineUp incomplete")
game_elm = f24_root.find("Game")
periods = [
Period(
id=1,
start_timestamp=None,
end_timestamp=None,
),
Period(
id=2,
start_timestamp=None,
end_timestamp=None,
),
]
possession_team = None
events = []
for event_elm in game_elm.iterchildren("Event"):
event_id = event_elm.attrib["id"]
type_id = int(event_elm.attrib["type_id"])
timestamp = _parse_f24_datetime(event_elm.attrib["timestamp"])
period_id = int(event_elm.attrib["period_id"])
for period in periods:
if period.id == period_id:
break
else:
logger.debug(
f"Skipping event {event_id} because period doesn't match {period_id}"
)
continue
if type_id == EVENT_TYPE_START_PERIOD:
logger.debug(
f"Set start of period {period.id} to {timestamp}")
period.start_timestamp = timestamp
elif type_id == EVENT_TYPE_END_PERIOD:
logger.debug(
f"Set end of period {period.id} to {timestamp}")
period.end_timestamp = timestamp
else:
if not period.start_timestamp:
# not started yet
continue
if event_elm.attrib["team_id"] == home_team.team_id:
team = teams[0]
elif event_elm.attrib["team_id"] == away_team.team_id:
team = teams[1]
else:
raise DeserializationError(
f"Unknown team_id {event_elm.attrib['team_id']}")
x = float(event_elm.attrib["x"])
y = float(event_elm.attrib["y"])
outcome = int(event_elm.attrib["outcome"])
raw_qualifiers = {
int(qualifier_elm.attrib["qualifier_id"]):
qualifier_elm.attrib.get("value")
for qualifier_elm in event_elm.iterchildren("Q")
}
player = None
if "player_id" in event_elm.attrib:
player = team.get_player_by_id(
event_elm.attrib["player_id"])
if type_id in BALL_OWNING_EVENTS:
possession_team = team
generic_event_kwargs = dict(
# from DataRecord
period=period,
timestamp=timestamp - period.start_timestamp,
ball_owning_team=possession_team,
ball_state=BallState.ALIVE,
# from Event
event_id=event_id,
team=team,
player=player,
coordinates=Point(x=x, y=y),
raw_event=event_elm,
)
if type_id == EVENT_TYPE_PASS:
pass_event_kwargs = _parse_pass(
raw_qualifiers, outcome)
event = PassEvent.create(
**pass_event_kwargs,
**generic_event_kwargs,
)
elif type_id == EVENT_TYPE_OFFSIDE_PASS:
pass_event_kwargs = _parse_offside_pass(raw_qualifiers)
event = PassEvent.create(
**pass_event_kwargs,
**generic_event_kwargs,
)
elif type_id == EVENT_TYPE_TAKE_ON:
take_on_event_kwargs = _parse_take_on(outcome)
event = TakeOnEvent.create(
qualifiers=None,
**take_on_event_kwargs,
**generic_event_kwargs,
)
elif type_id in (
EVENT_TYPE_SHOT_MISS,
EVENT_TYPE_SHOT_POST,
EVENT_TYPE_SHOT_SAVED,
EVENT_TYPE_SHOT_GOAL,
):
if type_id == EVENT_TYPE_SHOT_GOAL:
if 374 in raw_qualifiers.keys():
generic_event_kwargs["timestamp"] = (
_parse_f24_datetime(
raw_qualifiers.get(374).replace(
" ", "T")) -
period.start_timestamp)
shot_event_kwargs = _parse_shot(
raw_qualifiers,
type_id,
coordinates=generic_event_kwargs["coordinates"],
)
kwargs = {}
kwargs.update(generic_event_kwargs)
kwargs.update(shot_event_kwargs)
event = ShotEvent.create(**kwargs)
elif type_id == EVENT_TYPE_RECOVERY:
event = RecoveryEvent.create(
result=None,
qualifiers=None,
**generic_event_kwargs,
)
elif type_id == EVENT_TYPE_FOUL_COMMITTED:
event = FoulCommittedEvent.create(
result=None,
qualifiers=None,
**generic_event_kwargs,
)
elif type_id in BALL_OUT_EVENTS:
generic_event_kwargs["ball_state"] = BallState.DEAD
event = BallOutEvent.create(
result=None,
qualifiers=None,
**generic_event_kwargs,
)
elif type_id == EVENT_TYPE_FORMATION_CHANGE:
formation_change_event_kwargs = (
_parse_formation_change(raw_qualifiers))
event = FormationChangeEvent.create(
result=None,
qualifiers=None,
**formation_change_event_kwargs,
**generic_event_kwargs,
)
elif type_id == EVENT_TYPE_CARD:
generic_event_kwargs["ball_state"] = BallState.DEAD
card_event_kwargs = _parse_card(raw_qualifiers)
event = CardEvent.create(
**card_event_kwargs,
**generic_event_kwargs,
)
else:
event = GenericEvent.create(
**generic_event_kwargs,
result=None,
qualifiers=None,
event_name=_get_event_type_name(type_id),
)
if self.should_include_event(event):
events.append(transformer.transform_event(event))
metadata = Metadata(
teams=teams,
periods=periods,
pitch_dimensions=transformer.get_to_coordinate_system().
pitch_dimensions,
score=score,
frame_rate=None,
orientation=Orientation.ACTION_EXECUTING_TEAM,
flags=DatasetFlag.BALL_OWNING_TEAM,
provider=Provider.OPTA,
coordinate_system=transformer.get_to_coordinate_system(),
)
return EventDataset(
metadata=metadata,
records=events,
)
def deserialize(self, inputs: SportecInputs) -> EventDataset:
with performance_logging("load data", logger=logger):
match_root = objectify.fromstring(inputs.meta_data.read())
event_root = objectify.fromstring(inputs.event_data.read())
with performance_logging("parse data", logger=logger):
x_max = float(
match_root.MatchInformation.Environment.attrib["PitchX"]
)
y_max = float(
match_root.MatchInformation.Environment.attrib["PitchY"]
)
transformer = self.get_transformer(length=x_max, width=y_max)
team_path = objectify.ObjectPath(
"PutDataRequest.MatchInformation.Teams"
)
team_elms = list(team_path.find(match_root).iterchildren("Team"))
for team_elm in team_elms:
if team_elm.attrib["Role"] == "home":
home_team = _team_from_xml_elm(team_elm)
elif team_elm.attrib["Role"] == "guest":
away_team = _team_from_xml_elm(team_elm)
else:
raise DeserializationError(
f"Unknown side: {team_elm.attrib['Role']}"
)
(
home_score,
away_score,
) = match_root.MatchInformation.General.attrib["Result"].split(":")
score = Score(home=int(home_score), away=int(away_score))
teams = [home_team, away_team]
if len(home_team.players) == 0 or len(away_team.players) == 0:
raise DeserializationError("LineUp incomplete")
periods = []
period_id = 0
events = []
for event_elm in event_root.iterchildren("Event"):
event_chain = _event_chain_from_xml_elm(event_elm)
timestamp = _parse_datetime(event_chain["Event"]["EventTime"])
if (
SPORTEC_EVENT_NAME_KICKOFF in event_chain
and "GameSection"
in event_chain[SPORTEC_EVENT_NAME_KICKOFF]
):
period_id += 1
period = Period(
id=period_id,
start_timestamp=timestamp,
end_timestamp=None,
)
if period_id == 1:
team_left = event_chain[SPORTEC_EVENT_NAME_KICKOFF][
"TeamLeft"
]
if team_left == home_team.team_id:
# goal of home team is on the left side.
# this means they attack from left to right
orientation = Orientation.FIXED_HOME_AWAY
period.set_attacking_direction(
AttackingDirection.HOME_AWAY
)
else:
orientation = Orientation.FIXED_AWAY_HOME
period.set_attacking_direction(
AttackingDirection.AWAY_HOME
)
else:
last_period = periods[-1]
period.set_attacking_direction(
AttackingDirection.AWAY_HOME
if last_period.attacking_direction
== AttackingDirection.HOME_AWAY
else AttackingDirection.HOME_AWAY
)
periods.append(period)
elif SPORTEC_EVENT_NAME_FINAL_WHISTLE in event_chain:
period.end_timestamp = timestamp
continue
team = None
player = None
flatten_attributes = dict()
# reverse because top levels are more important
for event_attributes in reversed(event_chain.values()):
flatten_attributes.update(event_attributes)
if "Team" in flatten_attributes:
team = (
home_team
if flatten_attributes["Team"] == home_team.team_id
else away_team
)
if "Player" in flatten_attributes:
if not team:
raise ValueError("Player set while team is not set")
player = team.get_player_by_id(
flatten_attributes["Player"]
)
generic_event_kwargs = dict(
# from DataRecord
period=period,
timestamp=timestamp - period.start_timestamp,
ball_owning_team=None,
ball_state=BallState.ALIVE,
# from Event
event_id=event_chain["Event"]["EventId"],
coordinates=_parse_coordinates(event_chain["Event"]),
raw_event=flatten_attributes,
team=team,
player=player,
)
event_name, event_attributes = event_chain.popitem()
if event_name in SPORTEC_SHOT_EVENT_NAMES:
shot_event_kwargs = _parse_shot(
event_name=event_name, event_chain=event_chain
)
event = ShotEvent.create(
**shot_event_kwargs,
**generic_event_kwargs,
)
elif event_name in SPORTEC_PASS_EVENT_NAMES:
pass_event_kwargs = _parse_pass(
event_chain=event_chain, team=team
)
event = PassEvent.create(
**pass_event_kwargs,
**generic_event_kwargs,
receive_timestamp=None,
receiver_coordinates=None,
)
elif event_name == SPORTEC_EVENT_NAME_BALL_CLAIMING:
event = RecoveryEvent.create(
result=None,
qualifiers=None,
**generic_event_kwargs,
)
elif event_name == SPORTEC_EVENT_NAME_SUBSTITUTION:
substitution_event_kwargs = _parse_substitution(
event_attributes=event_attributes, team=team
)
generic_event_kwargs["player"] = substitution_event_kwargs[
"player"
]
del substitution_event_kwargs["player"]
event = SubstitutionEvent.create(
result=None,
qualifiers=None,
**substitution_event_kwargs,
**generic_event_kwargs,
)
elif event_name == SPORTEC_EVENT_NAME_CAUTION:
card_kwargs = _parse_caution(event_attributes)
event = CardEvent.create(
result=None,
qualifiers=None,
**card_kwargs,
**generic_event_kwargs,
)
elif event_name == SPORTEC_EVENT_NAME_FOUL:
foul_kwargs = _parse_foul(event_attributes, teams=teams)
generic_event_kwargs.update(foul_kwargs)
event = FoulCommittedEvent.create(
result=None,
qualifiers=None,
**generic_event_kwargs,
)
else:
event = GenericEvent.create(
result=None,
qualifiers=None,
event_name=event_name,
**generic_event_kwargs,
)
if events:
previous_event = events[-1]
if (
previous_event.event_type == EventType.PASS
and previous_event.result == PassResult.COMPLETE
):
if "X-Source-Position" in event_chain["Event"]:
previous_event.receiver_coordinates = Point(
x=float(
event_chain["Event"]["X-Source-Position"]
),
y=float(
event_chain["Event"]["Y-Source-Position"]
),
)
if (
event.event_type == EventType.PASS
and event.get_qualifier_value(SetPieceQualifier)
in (
SetPieceType.THROW_IN,
SetPieceType.GOAL_KICK,
SetPieceType.CORNER_KICK,
)
):
# 1. update previous pass
if events[-1].event_type == EventType.PASS:
events[-1].result = PassResult.OUT
# 2. add synthetic out event
decision_timestamp = _parse_datetime(
event_chain[list(event_chain.keys())[1]][
"DecisionTimestamp"
]
)
out_event = BallOutEvent.create(
period=period,
timestamp=decision_timestamp - period.start_timestamp,
ball_owning_team=None,
ball_state=BallState.DEAD,
# from Event
event_id=event_chain["Event"]["EventId"] + "-ball-out",
team=events[-1].team,
player=events[-1].player,
coordinates=None,
raw_event={},
result=None,
qualifiers=None,
)
events.append(transformer.transform_event(out_event))
events.append(transformer.transform_event(event))
events = list(
filter(
self.should_include_event,
events,
)
)
metadata = Metadata(
teams=teams,
periods=periods,
pitch_dimensions=transformer.get_to_coordinate_system().pitch_dimensions,
score=score,
frame_rate=None,
orientation=orientation,
flags=~(DatasetFlag.BALL_STATE | DatasetFlag.BALL_OWNING_TEAM),
provider=Provider.SPORTEC,
coordinate_system=transformer.get_to_coordinate_system(),
)
return EventDataset(
metadata=metadata,
records=events,
)