示例#1
0
    def __create_iterator(self, data: Readable, sample_rate: float,
                          frame_rate: int) -> 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 = columns[3]
            elif i == 1:
                player_jersey_numbers = columns[3:-2:2]
            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,
                        player_positions={
                            player_no: Point(
                                x=float(columns[3 + i * 2]),
                                y=1 - float(columns[3 + i * 2 + 1]),
                            )
                            for i, player_no in enumerate(
                                player_jersey_numbers)
                            if columns[3 + i * 2] != "NaN"
                        },
                        ball_position=Point(x=float(columns[-2]),
                                            y=1 - float(columns[-1]))
                        if columns[-2] != "NaN" else None,
                    )
                frame_idx += 1
示例#2
0
    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,
        )
示例#3
0
    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,)
示例#4
0
    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)
示例#5
0
    def deserialize(self,
                    inputs: Dict[str, Readable],
                    options: Dict = None) -> EventDataSet:
        self.__validate_inputs(inputs)

        periods = []
        period = None

        events = []

        game_state = self.__GameState(ball_state=BallState.DEAD,
                                      ball_owning_team=None)

        reader = csv.DictReader(
            map(lambda x: x.decode('utf-8'), inputs['raw_data']))
        for event_id, record in enumerate(reader):
            event_type = event_type_map[record['Type']]
            subtypes = record['Subtype'].split('-')

            start_timestamp = float(record['Start Time [s]'])
            end_timestamp = float(record['End Time [s]'])

            period_id = int(record['Period'])
            if not period or period.id != period_id:
                period = Period(id=period_id,
                                start_timestamp=start_timestamp,
                                end_timestamp=end_timestamp)
                periods.append(period)
            else:
                period.end_timestamp = end_timestamp

            if record['Team'] == 'Home':
                team = Team.HOME
            elif record['Team'] == 'Away':
                team = Team.AWAY
            else:
                raise ValueError(f'Unknown team: {record["team"]}')

            event_kwargs = dict(
                # From DataRecord:
                timestamp=start_timestamp,
                ball_owning_team=None,  ## todo
                ball_state=None,  # todo
                period=period,

                # From Event:
                event_id=event_id,
                team=team,
                end_timestamp=end_timestamp,
                player_jersey_no=record['From'][6:],
                position=Point(x=float(record['Start X']),
                               y=1 - float(record['Start Y']))
                if record['Start X'] != 'NaN' else None,
            )

            secondary_position = None
            if record['End X'] != 'NaN':
                secondary_position = Point(x=float(record['End X']),
                                           y=1 - float(record['End Y']))

            secondary_jersey_no = None
            if record['To']:
                secondary_jersey_no = record['To'][6:]

            event = None
            if event_type == EventType.SET_PIECE:
                set_piece, fk_attempt, retaken = \
                    build_subtypes(subtypes, [SetPiece, FKAttempt, Retaken])

                event = SetPieceEvent(**event_kwargs)
            elif event_type == EventType.RECOVERY:
                interference1, interference2 = \
                    build_subtypes(subtypes, [Interference1, Interference2])

                event = RecoveryEvent(**event_kwargs)
            elif event_type == EventType.PASS:
                body_part, attempt, deflection, offside = \
                    build_subtypes(subtypes, [BodyPart, Attempt, Deflection, Offside])

                event = PassEvent(
                    receiver_position=secondary_position,
                    receiver_player_jersey_no=secondary_jersey_no,
                    **event_kwargs)
            elif event_type == EventType.BALL_LOST:
                body_part, attempt, interference1, intervention, deflection, offside = \
                    build_subtypes(subtypes, [
                        BodyPart, Attempt, Interference1, Intervention,
                        Deflection, Offside
                    ])

                event = BallLossEvent(**event_kwargs)
            elif event_type == EventType.BALL_OUT:
                body_part, attempt, intervention, deflection, offside, own_goal = \
                    build_subtypes(subtypes, [
                        BodyPart, Attempt, Intervention, Deflection,
                        Offside, OwnGoal
                    ])

                event = BallOutEvent(**event_kwargs)
            elif event_type == EventType.SHOT:
                body_part, deflection, shot_direction, shot_result, offside = \
                    build_subtypes(subtypes, [
                        BodyPart, Deflection, ShotDirection,
                        ShotResult, Offside
                    ])

                event = ShotEvent(shot_result=shot_result, **event_kwargs)
            elif event_type == EventType.FAULT_RECEIVED:
                event = FaultReceivedEvent(**event_kwargs)
            elif event_type == EventType.CHALLENGE:
                challenge, fault, challenge_result = \
                    build_subtypes(subtypes, [Challenge, Fault, ChallengeResult])

                event = ChallengeEvent(**event_kwargs)
            elif event_type == EventType.CARD:
                card, = build_subtypes(subtypes, [Card])

                event = CardEvent(**event_kwargs)
            else:
                raise NotImplementedError(
                    f"EventType {event_type} not implemented")

            # We want to attach the game_state after the event to the event
            game_state = self.__reduce_game_state(event=event,
                                                  game_state=game_state)

            event.ball_state = game_state.ball_state
            event.ball_owning_team = game_state.ball_owning_team

            events.append(event)

        orientation = (
            Orientation.FIXED_HOME_AWAY if periods[0].attacking_direction
            == AttackingDirection.HOME_AWAY else Orientation.FIXED_AWAY_HOME)

        return EventDataSet(
            flags=DataSetFlag.BALL_STATE | DataSetFlag.BALL_OWNING_TEAM,
            orientation=orientation,
            pitch_dimensions=PitchDimensions(x_dim=Dimension(0, 1),
                                             y_dim=Dimension(0, 1)),
            periods=periods,
            records=events)
示例#6
0
    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,
        )
示例#7
0
    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,
        )