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"))
away_player_map = {}
home_player_map = {}
home_team_id = None
away_team_id = None
for team_elm in team_elms:
player_map = {
player_elm.attrib["PlayerRef"].lstrip("p"):
player_elm.attrib["ShirtNumber"]
for player_elm in team_elm.find(
"PlayerLineUp").iterchildren("MatchPlayer")
}
team_id = team_elm.attrib["TeamRef"].lstrip("t")
if team_elm.attrib["Side"] == "Home":
home_player_map = player_map
home_team_id = team_id
elif team_elm.attrib["Side"] == "Away":
away_player_map = player_map
away_team_id = team_id
else:
raise Exception(f"Unknown side: {team_elm.attrib['Side']}")
if not away_player_map or not home_player_map:
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,
),
]
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_id:
team = Team.HOME
current_team_map = home_player_map
elif event_elm.attrib["team_id"] == away_team_id:
team = Team.AWAY
current_team_map = away_player_map
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_jersey_no = None
if "player_id" in event_elm.attrib:
player_jersey_no = current_team_map[
event_elm.attrib["player_id"]]
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_id,
team=team,
player_jersey_no=player_jersey_no,
position=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,
position=generic_event_kwargs["position"],
)
kwargs = {}
kwargs.update(generic_event_kwargs)
kwargs.update(shot_event_kwargs)
event = ShotEvent(**kwargs)
else:
event = GenericEvent(**generic_event_kwargs,
result=None)
if (not wanted_event_types
or event.event_type in wanted_event_types):
events.append(event)
return EventDataset(
flags=DatasetFlag.BALL_OWNING_TEAM,
orientation=Orientation.ACTION_EXECUTING_TEAM,
pitch_dimensions=PitchDimensions(x_dim=Dimension(0, 100),
y_dim=Dimension(0, 100)),
periods=periods,
records=events,
)
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_player_map = {
player['player_id']: str(player['jersey_number'])
for player in home_lineup['lineup']
}
away_player_map = {
player['player_id']: str(player['jersey_number'])
for player in away_lineup['lineup']
}
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 = Team.HOME
current_team_map = home_player_map
elif raw_event['team']['id'] == away_lineup['team_id']:
team = Team.AWAY
current_team_map = away_player_map
else:
raise Exception(
f"Unknown team_id {raw_event['team']['id']}")
if raw_event['possession_team']['id'] == home_lineup[
'team_id']:
possession_team = Team.HOME
elif raw_event['possession_team']['id'] == away_lineup[
'team_id']:
possession_team = Team.AWAY
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
else timestamp + period.end_timestamp,
end_timestamp=None)
periods.append(period)
else:
period.end_timestamp = period.start_timestamp + timestamp
player_jersey_no = None
if 'player' in raw_event:
player_jersey_no = current_team_map[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_jersey_no=player_jersey_no,
position=(_parse_position(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'],
current_team_map=current_team_map,
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, **generic_event_kwargs)
if not wanted_event_types or event.event_type in wanted_event_types:
events.append(event)
return EventDataset(flags=DatasetFlag.BALL_OWNING_TEAM,
orientation=Orientation.ACTION_EXECUTING_TEAM,
pitch_dimensions=PitchDimensions(
x_dim=Dimension(0, 120),
y_dim=Dimension(0, 80)),
periods=periods,
records=events)
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 deserialize(self,
inputs: Dict[str, Readable],
options: Dict = None) -> EventDataset:
"""
Deserialize Metrica Sports event data json format into a `EventDataset`.
Parameters
----------
inputs : dict
input `raw_data` should point to a `Readable` object containing
the 'json' formatted event data. input `metadata` should point
to a `Readable` object containing the `xml` metadata file.
options : dict
Options for deserialization of the Metrica Sports event json 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 = MetricaEventsJsonSerializer()
>>> with open("events.json", "rb") as raw_data, \
>>> open("metadata.xml", "rb") as metadata:
>>>
>>> dataset = serializer.deserialize(
>>> inputs={
>>> 'raw_data': raw_data,
>>> 'metadata': metadata
>>> },
>>> 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["raw_data"])
metadata = load_metadata(inputs["metadata"],
provider=Provider.METRICA)
with performance_logging("parse data", logger=logger):
wanted_event_types = [
EventType[event_type.upper()]
for event_type in options.get("event_types", [])
]
events = []
for raw_event in raw_events["data"]:
if raw_event["team"]["id"] == metadata.teams[0].team_id:
team = metadata.teams[0]
elif raw_event["team"]["id"] == metadata.teams[1].team_id:
team = metadata.teams[1]
else:
raise Exception(
f"Unknown team_id {raw_event['team']['id']}")
player = team.get_player_by_id(raw_event["from"]["id"])
event_type = raw_event["type"]["id"]
subtypes = _parse_subtypes(raw_event)
period = [
period for period in metadata.periods
if period.id == raw_event["period"]
][0]
generic_event_kwargs = dict(
# from DataRecord
period=period,
timestamp=raw_event["start"]["time"],
ball_owning_team=_parse_ball_owning_team(event_type, team),
ball_state=BallState.ALIVE,
# from Event
event_id=None,
team=team,
player=player,
coordinates=(_parse_coordinates(raw_event["start"])),
raw_event=raw_event,
)
if event_type in MS_PASS_TYPES:
pass_event_kwargs = _parse_pass(
event=raw_event,
subtypes=subtypes,
team=team,
)
event = PassEvent(
**pass_event_kwargs,
**generic_event_kwargs,
)
elif event_type == MS_EVENT_TYPE_SHOT:
shot_event_kwargs = _parse_shot(event=raw_event,
subtypes=subtypes)
event = ShotEvent(**shot_event_kwargs,
**generic_event_kwargs)
elif subtypes and MS_EVENT_TYPE_DRIBBLE in subtypes:
take_on_event_kwargs = _parse_take_on(subtypes=subtypes)
event = TakeOnEvent(**take_on_event_kwargs,
**generic_event_kwargs)
elif event_type == MS_EVENT_TYPE_CARRY:
carry_event_kwargs = _parse_carry(event=raw_event, )
event = CarryEvent(
**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)
return EventDataset(
metadata=metadata,
records=events,
)