def _set_skillcorner_attacking_directions(cls, frames, periods):
"""
with only partial tracking data we cannot rely on a single frame to
infer the attacking directions as a simple average of only some players
x-coords might not reflect the attacking direction.
"""
attacking_directions = []
for frame in frames:
if len(frame.players_data) > 0:
attacking_directions.append(
attacking_direction_from_frame(frame))
else:
attacking_directions.append(AttackingDirection.NOT_SET)
frame_periods = np.array([_frame.period.id for _frame in frames])
for period in periods.keys():
if period in frame_periods:
count = Counter(
np.array(attacking_directions)[frame_periods == period])
att_direction = count.most_common()[0][0]
periods[period].attacking_direction = att_direction
else:
periods[
period].attacking_direction = AttackingDirection.NOT_SET
def deserialize(self,
inputs: Dict[str, Readable],
options: Dict = None) -> TrackingDataset:
"""
Deserialize Metrica tracking data into a `TrackingDataset`.
Parameters
----------
inputs : dict
input `raw_data_home` should point to a `Readable` object containing
the 'csv' formatted raw data for the home team. input `raw_data_away` should point
to a `Readable` object containing the 'csv' formatted raw data for the away team.
options : dict
Options for deserialization of the Metrica file. Possible options are
`sample_rate` (float between 0 and 1) to specify the amount of
frames that should be loaded, `limit` to specify the max number of
frames that will be returned.
Returns
-------
dataset : TrackingDataset
Raises
------
ValueError when both input files don't seem to belong to each other
See Also
--------
Examples
--------
>>> serializer = MetricaTrackingSerializer()
>>> with open("Sample_Game_1_RawTrackingData_Away_Team.csv", "rb") as raw_home, \
>>> open("Sample_Game_1_RawTrackingData_Home_Team.csv", "rb") as raw_away:
>>>
>>> dataset = serializer.deserialize(
>>> inputs={
>>> 'raw_data_home': raw_home,
>>> 'raw_data_away': raw_away
>>> },
>>> options={
>>> 'sample_rate': 1/12
>>> }
>>> )
"""
self.__validate_inputs(inputs)
if not options:
options = {}
sample_rate = float(options.get('sample_rate', 1.0))
limit = int(options.get('limit', 0))
# consider reading this from data
frame_rate = 25
with performance_logging("prepare", logger=logger):
home_iterator = self.__create_iterator(inputs['raw_data_home'],
sample_rate, frame_rate)
away_iterator = self.__create_iterator(inputs['raw_data_away'],
sample_rate, frame_rate)
partial_frames = zip(home_iterator, away_iterator)
with performance_logging("loading", logger=logger):
frames = []
periods = []
partial_frame_type = self.__PartialFrame
home_partial_frame: partial_frame_type
away_partial_frame: partial_frame_type
for n, (home_partial_frame,
away_partial_frame) in enumerate(partial_frames):
self.__validate_partials(home_partial_frame,
away_partial_frame)
period: Period = home_partial_frame.period
frame_id: int = home_partial_frame.frame_id
frame = Frame(frame_id=frame_id,
timestamp=frame_id / frame_rate -
period.start_timestamp,
ball_position=home_partial_frame.ball_position,
home_team_player_positions=home_partial_frame.
player_positions,
away_team_player_positions=away_partial_frame.
player_positions,
period=period,
ball_state=None,
ball_owning_team=None)
frames.append(frame)
if not periods or period.id != periods[-1].id:
periods.append(period)
if not period.attacking_direction_set:
period.set_attacking_direction(
attacking_direction=attacking_direction_from_frame(
frame))
n += 1
if limit and n >= limit:
break
orientation = (
Orientation.FIXED_HOME_AWAY if periods[0].attacking_direction
== AttackingDirection.HOME_AWAY else Orientation.FIXED_AWAY_HOME)
return TrackingDataset(
flags=~(DatasetFlag.BALL_STATE | DatasetFlag.BALL_OWNING_TEAM),
frame_rate=frame_rate,
orientation=orientation,
pitch_dimensions=PitchDimensions(x_dim=Dimension(0, 1),
y_dim=Dimension(0, 1)),
periods=periods,
records=frames)
def deserialize(self, inputs: TRACABInputs) -> TrackingDataset:
# TODO: also used in Metrica, extract to a method
home_team = Team(team_id="home", name="home", ground=Ground.HOME)
away_team = Team(team_id="away", name="away", ground=Ground.AWAY)
teams = [home_team, away_team]
with performance_logging("Loading metadata", logger=logger):
match = objectify.fromstring(inputs.meta_data.read()).match
frame_rate = int(match.attrib["iFrameRateFps"])
pitch_size_width = float(match.attrib["fPitchXSizeMeters"])
pitch_size_height = float(match.attrib["fPitchYSizeMeters"])
periods = []
for period in match.iterchildren(tag="period"):
start_frame_id = int(period.attrib["iStartFrame"])
end_frame_id = int(period.attrib["iEndFrame"])
if start_frame_id != 0 or end_frame_id != 0:
periods.append(
Period(
id=int(period.attrib["iId"]),
start_timestamp=start_frame_id / frame_rate,
end_timestamp=end_frame_id / frame_rate,
)
)
with performance_logging("Loading data", logger=logger):
transformer = self.get_transformer(
length=pitch_size_width, width=pitch_size_height
)
def _iter():
n = 0
sample = 1.0 / self.sample_rate
for line_ in inputs.raw_data.readlines():
line_ = line_.strip().decode("ascii")
if not line_:
continue
frame_id = int(line_[:10].split(":", 1)[0])
if self.only_alive and not line_.endswith("Alive;:"):
continue
for period_ in periods:
if period_.contains(frame_id / frame_rate):
if n % sample == 0:
yield period_, line_
n += 1
frames = []
for n, (period, line) in enumerate(_iter()):
frame = self._frame_from_line(teams, period, line, frame_rate)
frame = transformer.transform_frame(frame)
frames.append(frame)
if not period.attacking_direction_set:
period.set_attacking_direction(
attacking_direction=attacking_direction_from_frame(
frame
)
)
if self.limit and n >= self.limit:
break
orientation = (
Orientation.FIXED_HOME_AWAY
if periods[0].attacking_direction == AttackingDirection.HOME_AWAY
else Orientation.FIXED_AWAY_HOME
)
metadata = Metadata(
teams=teams,
periods=periods,
pitch_dimensions=transformer.get_to_coordinate_system().pitch_dimensions,
score=None,
frame_rate=frame_rate,
orientation=orientation,
provider=Provider.TRACAB,
flags=DatasetFlag.BALL_OWNING_TEAM | DatasetFlag.BALL_STATE,
coordinate_system=transformer.get_to_coordinate_system(),
)
return TrackingDataset(
records=frames,
metadata=metadata,
)
def deserialize(self,
inputs: Dict[str, Readable],
options: Dict = None) -> TrackingDataSet:
"""
Deserialize TRACAB tracking data into a `TrackingDataSet`.
Parameters
----------
inputs : dict
input `raw_data` should point to a `Readable` object containing
the 'csv' formatted raw data. input `meta_data` should point to
the xml metadata data.
options : dict
Options for deserialization of the TRACAB file. Possible options are
`only_alive` (boolean) to specify that only frames with alive ball state
should be loaded, or `sample_rate` (float between 0 and 1) to specify
the amount of frames that should be loaded.
Returns
-------
data_set : TrackingDataSet
Raises
------
-
See Also
--------
Examples
--------
>>> serializer = TRACABSerializer()
>>> with open("metadata.xml", "rb") as meta, \
>>> open("raw.dat", "rb") as raw:
>>> data_set = serializer.deserialize(
>>> inputs={
>>> 'meta_data': meta,
>>> 'raw_data': raw
>>> },
>>> options={
>>> 'only_alive': True,
>>> 'sample_rate': 1/12
>>> }
>>> )
"""
self.__validate_inputs(inputs)
if not options:
options = {}
sample_rate = float(options.get('sample_rate', 1.0))
only_alive = bool(options.get('only_alive', True))
with performance_logging("Loading metadata"):
match = objectify.fromstring(inputs['meta_data'].read()).match
frame_rate = int(match.attrib['iFrameRateFps'])
pitch_size_width = float(match.attrib['fPitchXSizeMeters'])
pitch_size_height = float(match.attrib['fPitchYSizeMeters'])
periods = []
for period in match.iterchildren(tag='period'):
start_frame_id = int(period.attrib['iStartFrame'])
end_frame_id = int(period.attrib['iEndFrame'])
if start_frame_id != 0 or end_frame_id != 0:
periods.append(
Period(id=int(period.attrib['iId']),
start_timestamp=start_frame_id / frame_rate,
end_timestamp=end_frame_id / frame_rate))
with performance_logging("Loading data"):
def _iter():
n = 0
sample = 1. / sample_rate
for line in inputs['raw_data'].readlines():
line = line.strip().decode("ascii")
if not line:
continue
frame_id = int(line[:10].split(":", 1)[0])
if only_alive and not line.endswith("Alive;:"):
continue
for period in periods:
if period.contains(frame_id / frame_rate):
if n % sample == 0:
yield period, line
n += 1
frames = []
for period, line in _iter():
frame = self._frame_from_line(period, line, frame_rate)
frames.append(frame)
if not period.attacking_direction_set:
period.set_attacking_direction(
attacking_direction=attacking_direction_from_frame(
frame))
orientation = (
Orientation.FIXED_HOME_AWAY if periods[0].attacking_direction
== AttackingDirection.HOME_AWAY else Orientation.FIXED_AWAY_HOME)
return TrackingDataSet(flags=DataSetFlag.BALL_OWNING_TEAM
| DataSetFlag.BALL_STATE,
frame_rate=frame_rate,
orientation=orientation,
pitch_dimensions=PitchDimensions(
x_dim=Dimension(-1 * pitch_size_width / 2,
pitch_size_width / 2),
y_dim=Dimension(-1 * pitch_size_height / 2,
pitch_size_height / 2),
x_per_meter=100,
y_per_meter=100),
periods=periods,
records=frames)
def deserialize(self,
inputs: Dict[str, Readable],
options: Dict = None) -> TrackingDataset:
"""
Deserialize EPTS tracking data into a `TrackingDataset`.
Parameters
----------
inputs : dict
input `raw_data` should point to a `Readable` object containing
the 'csv' formatted raw data. input `meta_data` should point to
the xml metadata data.
options : dict
Options for deserialization of the EPTS file. Possible options are
`sample_rate` (float between 0 and 1) to specify the amount of
frames that should be loaded, `limit` to specify the max number of
frames that will be returned.
Returns
-------
dataset : TrackingDataset
Raises
------
-
See Also
--------
Examples
--------
>>> serializer = EPTSSerializer()
>>> with open("metadata.xml", "rb") as meta, \
>>> open("raw.dat", "rb") as raw:
>>> dataset = serializer.deserialize(
>>> inputs={
>>> 'meta_data': meta,
>>> 'raw_data': raw
>>> },
>>> options={
>>> 'sample_rate': 1/12
>>> }
>>> )
"""
self.__validate_inputs(inputs)
if not options:
options = {}
sample_rate = float(options.get("sample_rate", 1.0))
limit = int(options.get("limit", 0))
with performance_logging("Loading metadata", logger=logger):
meta_data = load_meta_data(inputs["meta_data"])
periods = meta_data.periods
with performance_logging("Loading data", logger=logger):
# assume they are sorted
frames = [
self._frame_from_row(row, meta_data) for row in read_raw_data(
raw_data=inputs["raw_data"],
meta_data=meta_data,
sensor_ids=["position"
], # we don't care about other sensors
sample_rate=sample_rate,
limit=limit,
)
]
if periods:
start_attacking_direction = periods[0].attacking_direction
elif frames:
start_attacking_direction = attacking_direction_from_frame(
frames[0])
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)
return TrackingDataset(
flags=~(DatasetFlag.BALL_STATE | DatasetFlag.BALL_OWNING_TEAM),
frame_rate=meta_data.frame_rate,
orientation=orientation,
pitch_dimensions=meta_data.pitch_dimensions,
periods=periods,
records=frames,
)
def deserialize(
self, inputs: Dict[str, Readable], options: Dict = None
) -> TrackingDataset:
"""
Deserialize TRACAB tracking data into a `TrackingDataset`.
Parameters
----------
inputs : dict
input `raw_data` should point to a `Readable` object containing
the 'csv' formatted raw data. input `metadata` should point to
the xml metadata data.
options : dict
Options for deserialization of the TRACAB file. Possible options are
`only_alive` (boolean) to specify that only frames with alive ball state
should be loaded, or `sample_rate` (float between 0 and 1) to specify
the amount of frames that should be loaded, `limit` to specify the max number of
frames that will be returned.
Returns
-------
dataset : TrackingDataset
Raises
------
-
See Also
--------
Examples
--------
>>> serializer = TRACABSerializer()
>>> with open("metadata.xml", "rb") as meta, \
>>> open("raw.dat", "rb") as raw:
>>> dataset = serializer.deserialize(
>>> inputs={
>>> 'metadata': meta,
>>> 'raw_data': raw
>>> },
>>> options={
>>> 'only_alive': True,
>>> 'sample_rate': 1/12
>>> }
>>> )
"""
self.__validate_inputs(inputs)
if not options:
options = {}
sample_rate = float(options.get("sample_rate", 1.0))
limit = int(options.get("limit", 0))
only_alive = bool(options.get("only_alive", True))
# TODO: also used in Metrica, extract to a method
home_team = Team(team_id="home", name="home", ground=Ground.HOME)
away_team = Team(team_id="away", name="away", ground=Ground.AWAY)
teams = [home_team, away_team]
with performance_logging("Loading metadata", logger=logger):
match = objectify.fromstring(inputs["metadata"].read()).match
frame_rate = int(match.attrib["iFrameRateFps"])
pitch_size_width = float(match.attrib["fPitchXSizeMeters"])
pitch_size_height = float(match.attrib["fPitchYSizeMeters"])
periods = []
for period in match.iterchildren(tag="period"):
start_frame_id = int(period.attrib["iStartFrame"])
end_frame_id = int(period.attrib["iEndFrame"])
if start_frame_id != 0 or end_frame_id != 0:
periods.append(
Period(
id=int(period.attrib["iId"]),
start_timestamp=start_frame_id / frame_rate,
end_timestamp=end_frame_id / frame_rate,
)
)
with performance_logging("Loading data", logger=logger):
def _iter():
n = 0
sample = 1.0 / sample_rate
for line_ in inputs["raw_data"].readlines():
line_ = line_.strip().decode("ascii")
if not line_:
continue
frame_id = int(line_[:10].split(":", 1)[0])
if only_alive and not line_.endswith("Alive;:"):
continue
for period_ in periods:
if period_.contains(frame_id / frame_rate):
if n % sample == 0:
yield period_, line_
n += 1
frames = []
for n, (period, line) in enumerate(_iter()):
frame = self._frame_from_line(teams, period, line, frame_rate)
frames.append(frame)
if not period.attacking_direction_set:
period.set_attacking_direction(
attacking_direction=attacking_direction_from_frame(
frame
)
)
if limit and n >= limit:
break
orientation = (
Orientation.FIXED_HOME_AWAY
if periods[0].attacking_direction == AttackingDirection.HOME_AWAY
else Orientation.FIXED_AWAY_HOME
)
metadata = Metadata(
teams=teams,
periods=periods,
pitch_dimensions=PitchDimensions(
x_dim=Dimension(
-1 * pitch_size_width / 2, pitch_size_width / 2
),
y_dim=Dimension(
-1 * pitch_size_height / 2, pitch_size_height / 2
),
x_per_meter=100,
y_per_meter=100,
),
score=None,
frame_rate=frame_rate,
orientation=orientation,
provider=Provider.TRACAB,
flags=DatasetFlag.BALL_OWNING_TEAM | DatasetFlag.BALL_STATE,
)
return TrackingDataset(
records=frames,
metadata=metadata,
)
def deserialize(self, inputs: SecondSpectrumInputs) -> TrackingDataset:
metadata = None
# Handles the XML metadata that contains the pitch dimensions and frame info
with performance_logging("Loading XML metadata", logger=logger):
# The meta data can also be in JSON format. In that case
# it also contains the 'additional metadata'.
# First do a 'peek' to determine the char
first_byte = inputs.meta_data.read(1)
if first_byte == b"{":
metadata = json.loads(first_byte + inputs.meta_data.read())
frame_rate = int(metadata["fps"])
pitch_size_height = float(metadata["pitchLength"])
pitch_size_width = float(metadata["pitchWidth"])
periods = []
for period in metadata["periods"]:
start_frame_id = int(period["startFrameIdx"])
end_frame_id = int(period["endFrameIdx"])
if start_frame_id != 0 or end_frame_id != 0:
# Frame IDs are unix timestamps (in milliseconds)
periods.append(
Period(
id=int(period["number"]),
start_timestamp=start_frame_id,
end_timestamp=end_frame_id,
)
)
else:
match = objectify.fromstring(
first_byte + inputs.meta_data.read()
).match
frame_rate = int(match.attrib["iFrameRateFps"])
pitch_size_height = float(match.attrib["fPitchYSizeMeters"])
pitch_size_width = float(match.attrib["fPitchXSizeMeters"])
periods = []
for period in match.iterchildren(tag="period"):
start_frame_id = int(period.attrib["iStartFrame"])
end_frame_id = int(period.attrib["iEndFrame"])
if start_frame_id != 0 or end_frame_id != 0:
# Frame IDs are unix timestamps (in milliseconds)
periods.append(
Period(
id=int(period.attrib["iId"]),
start_timestamp=start_frame_id,
end_timestamp=end_frame_id,
)
)
# Default team initialisation
home_team = Team(team_id="home", name="home", ground=Ground.HOME)
away_team = Team(team_id="away", name="away", ground=Ground.AWAY)
teams = [home_team, away_team]
if inputs.additional_meta_data or metadata:
with performance_logging("Loading JSON metadata", logger=logger):
try:
if inputs.additional_meta_data:
metadata = json.loads(
inputs.additional_meta_data.read()
)
home_team_id = metadata["homeOptaId"]
away_team_id = metadata["awayOptaId"]
# Tries to parse (short) team names from the description string
try:
home_name = (
metadata["description"].split("-")[0].strip()
)
away_name = (
metadata["description"]
.split("-")[1]
.split(":")[0]
.strip()
)
except:
home_name, away_name = "home", "away"
teams[0].team_id = home_team_id
teams[0].name = home_name
teams[1].team_id = away_team_id
teams[1].name = away_name
for team, team_str in zip(
teams, ["homePlayers", "awayPlayers"]
):
for player_data in metadata[team_str]:
# We use the attributes field of Player to store the extra IDs provided by the
# metadata. We designate the player_id to be the 'optaId' field as this is what's
# used as 'player_id' in the raw frame data file
player_attributes = {
k: v
for k, v in player_data.items()
if k in ["ssiId", "optaUuid"]
}
player = Player(
player_id=player_data["optaId"],
name=player_data["name"],
starting=player_data["position"] != "SUB",
position=player_data["position"],
team=team,
jersey_no=int(player_data["number"]),
attributes=player_attributes,
)
team.players.append(player)
except: # TODO: More specific exception
logging.warning(
"Optional JSON Metadata is malformed. Continuing without"
)
# Handles the tracking frame data
with performance_logging("Loading data", logger=logger):
transformer = self.get_transformer(
length=pitch_size_width, width=pitch_size_height
)
def _iter():
n = 0
sample = 1 / self.sample_rate
for line_ in inputs.raw_data.readlines():
line_ = line_.strip().decode("ascii")
if not line_:
continue
# Each line is just json so we just parse it
frame_data = json.loads(line_)
if self.only_alive and not frame_data["live"]:
continue
if n % sample == 0:
yield frame_data
n += 1
frames = []
for n, frame_data in enumerate(_iter()):
period = periods[frame_data["period"] - 1]
frame = self._frame_from_framedata(teams, period, frame_data)
frame = transformer.transform_frame(frame)
frames.append(frame)
if not period.attacking_direction_set:
period.set_attacking_direction(
attacking_direction=attacking_direction_from_frame(
frame
)
)
if self.limit and n + 1 >= self.limit:
break
orientation = (
Orientation.FIXED_HOME_AWAY
if periods[0].attacking_direction == AttackingDirection.HOME_AWAY
else Orientation.FIXED_AWAY_HOME
)
metadata = Metadata(
teams=teams,
periods=periods,
pitch_dimensions=transformer.get_to_coordinate_system().pitch_dimensions,
score=None,
frame_rate=frame_rate,
orientation=orientation,
provider=Provider.SECONDSPECTRUM,
flags=DatasetFlag.BALL_OWNING_TEAM | DatasetFlag.BALL_STATE,
coordinate_system=transformer.get_to_coordinate_system(),
)
return TrackingDataset(
records=frames,
metadata=metadata,
)
