File size: 7,812 Bytes
d6ea71e |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 |
"""Implements the feature tranformers of the VAEP framework."""
from typing import Any, Callable, Union
import numpy as np
import pandas as pd
from pandera.typing import DataFrame
import socceraction.atomic.spadl.config as atomicspadl
from socceraction.atomic.spadl import AtomicSPADLSchema
from socceraction.spadl import SPADLSchema
from socceraction.vaep.features import (
actiontype,
bodypart,
bodypart_detailed,
bodypart_detailed_onehot,
bodypart_onehot,
gamestates,
player_possession_time,
simple,
speed,
team,
time,
time_delta,
)
__all__ = [
"feature_column_names",
"play_left_to_right",
"gamestates",
"actiontype",
"actiontype_onehot",
"bodypart",
"bodypart_detailed",
"bodypart_onehot",
"bodypart_detailed_onehot",
"team",
"time",
"time_delta",
"speed",
"location",
"polar",
"movement_polar",
"direction",
"goalscore",
"player_possession_time",
]
Actions = Union[DataFrame[SPADLSchema], DataFrame[AtomicSPADLSchema]]
GameStates = list[Actions]
Features = DataFrame[Any]
FeatureTransfomer = Callable[[GameStates], Features]
def feature_column_names(fs: list[FeatureTransfomer], nb_prev_actions: int = 3) -> list[str]:
"""Return the names of the features generated by a list of transformers.
Parameters
----------
fs : list(callable)
A list of feature transformers.
nb_prev_actions : int, default=3 # noqa: DAR103
The number of previous actions included in the game state.
Returns
-------
list(str)
The name of each generated feature.
"""
spadlcolumns = [
"game_id",
"original_event_id",
"action_id",
"period_id",
"time_seconds",
"team_id",
"player_id",
"x",
"y",
"dx",
"dy",
"bodypart_id",
"bodypart_name",
"type_id",
"type_name",
]
dummy_actions = pd.DataFrame(np.zeros((10, len(spadlcolumns))), columns=spadlcolumns)
for c in spadlcolumns:
if "name" in c:
dummy_actions[c] = dummy_actions[c].astype(str)
gs = gamestates(dummy_actions, nb_prev_actions) # type: ignore
return list(pd.concat([f(gs) for f in fs], axis=1).columns)
def play_left_to_right(gamestates: GameStates, home_team_id: int) -> GameStates:
"""Perform all action in the same playing direction.
This changes the start and end location of each action, such that all actions
are performed as if the team plays from left to right.
Parameters
----------
gamestates : GameStates
The game states of a game.
home_team_id : int
The ID of the home team.
Returns
-------
list(pd.DataFrame)
The game states with all actions performed left to right.
"""
a0 = gamestates[0]
away_idx = a0.team_id != home_team_id
for actions in gamestates:
actions.loc[away_idx, "x"] = atomicspadl.field_length - actions[away_idx]["x"].values
actions.loc[away_idx, "y"] = atomicspadl.field_width - actions[away_idx]["y"].values
actions.loc[away_idx, "dx"] = -actions[away_idx]["dx"].values
actions.loc[away_idx, "dy"] = -actions[away_idx]["dy"].values
return gamestates
@simple
def actiontype_onehot(actions: Actions) -> Features:
"""Get the one-hot-encoded type of each action.
Parameters
----------
actions : Actions
The actions of a game.
Returns
-------
Features
A one-hot encoding of each action's type.
"""
X = {}
for type_id, type_name in enumerate(atomicspadl.actiontypes):
col = "actiontype_" + type_name
X[col] = actions["type_id"] == type_id
return pd.DataFrame(X, index=actions.index)
@simple
def location(actions: Actions) -> Features:
"""Get the location where each action started.
Parameters
----------
actions : Actions
The actions of a game.
Returns
-------
Features
The 'x' and 'y' location of each action.
"""
return actions[["x", "y"]]
_goal_x = atomicspadl.field_length
_goal_y = atomicspadl.field_width / 2
@simple
def polar(actions: Actions) -> Features:
"""Get the polar coordinates of each action's start location.
The center of the opponent's goal is used as the origin.
Parameters
----------
actions : Actions
The actions of a game.
Returns
-------
Features
The 'dist_to_goal' and 'angle_to_goal' of each action.
"""
polardf = pd.DataFrame(index=actions.index)
dx = (_goal_x - actions["x"]).abs().values
dy = (_goal_y - actions["y"]).abs().values
polardf["dist_to_goal"] = np.sqrt(dx**2 + dy**2)
with np.errstate(divide="ignore", invalid="ignore"):
polardf["angle_to_goal"] = np.nan_to_num(np.arctan(dy / dx))
return polardf
@simple
def movement_polar(actions: Actions) -> Features:
"""Get the distance covered and direction of each action.
Parameters
----------
actions : Actions
The actions of a game.
Returns
-------
Features
The distance covered ('mov_d') and direction ('mov_angle') of each action.
"""
mov = pd.DataFrame(index=actions.index)
mov["mov_d"] = np.sqrt(actions.dx**2 + actions.dy**2)
with np.errstate(divide="ignore", invalid="ignore"):
mov["mov_angle"] = np.arctan2(actions.dy, actions.dx)
mov.loc[actions.dy == 0, "mov_angle"] = 0 # fix float errors
return mov
@simple
def direction(actions: Actions) -> Features:
"""Get the direction of the action as components of the unit vector.
Parameters
----------
actions : Actions
The actions of a game.
Returns
-------
Features
The x-component ('dx') and y-compoment ('mov_angle') of the unit
vector of each action.
"""
mov = pd.DataFrame(index=actions.index)
totald = np.sqrt(actions.dx**2 + actions.dy**2)
for d in ["dx", "dy"]:
# we don't want to give away the end location,
# just the direction of the ball
# We also don't want to divide by zero
mov[d] = actions[d].mask(totald > 0, actions[d] / totald)
return mov
def goalscore(gamestates: GameStates) -> Features:
"""Get the number of goals scored by each team after the action.
Parameters
----------
gamestates : GameStates
The gamestates of a game.
Returns
-------
Features
The number of goals scored by the team performing the last action of the
game state ('goalscore_team'), by the opponent ('goalscore_opponent'),
and the goal difference between both teams ('goalscore_diff').
"""
actions = gamestates[0]
teamA = actions["team_id"].values[0]
goals = actions.type_name == "goal"
owngoals = actions["type_name"].str.contains("owngoal")
teamisA = actions["team_id"] == teamA
teamisB = ~teamisA
goalsteamA = (goals & teamisA) | (owngoals & teamisB)
goalsteamB = (goals & teamisB) | (owngoals & teamisA)
goalscoreteamA = goalsteamA.cumsum() - goalsteamA
goalscoreteamB = goalsteamB.cumsum() - goalsteamB
scoredf = pd.DataFrame(index=actions.index)
scoredf["goalscore_team"] = (goalscoreteamA * teamisA) + (goalscoreteamB * teamisB)
scoredf["goalscore_opponent"] = (goalscoreteamB * teamisA) + (goalscoreteamA * teamisB)
scoredf["goalscore_diff"] = scoredf["goalscore_team"] - scoredf["goalscore_opponent"]
return scoredf
|