def test_one_on_one(self):
game_info = GlickoGameInfo()
player1 = Team({1: (1500, 200)})
player2 = Team({2: (1400, 30)})
player3 = Team({3: (1550, 100)})
player4 = Team({4: (1700, 300)})
matches = Matches([Match([player1, player2], [1, 2]),
Match([player1, player3], [2, 1]),
Match([player1, player4], [2, 1])])
new_ratings = self.calculator.new_ratings(matches, 1, game_info)
# self.assertMatchQuality(1.0, self.calculator.calculate_match_quality(matches, game_info))
self.assertRating(1464.1, 151.4, new_ratings.rating_by_id(1))
def updated_ratings(self):
results = Match()
for current_team in self.prior_layer.output_variables_groups:
team_results = Team()
for current_player, current_player_rating in [(player.key, player.value) for player in current_team]:
new_rating = GaussianRating(current_player_rating.mean,
current_player_rating.stdev)
team_results[current_player] = new_rating
results.append(team_results)
return results
def updated_ratings(self):
results = Match()
for current_team in self.prior_layer.output_variables_groups:
team_results = Team()
for current_player, current_player_rating in [
(player.key, player.value) for player in current_team
]:
new_rating = GaussianRating(current_player_rating.mean,
current_player_rating.stdev)
team_results[current_player] = new_rating
results.append(team_results)
return results
def test_two_on_four_on_two_win_draw(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (40.0, 4.0),
2: (45.0, 3.0)
}, {
3: (20.0, 7.0),
4: (19.0, 6.0),
5: (30.0, 9.0),
6: (10.0, 4.0)
}, {
7: (50.0, 5.0),
8: (30.0, 2.0)
}],
rank=[1, 2, 2])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.367, self.calculator.match_quality(teams, game_info))
self.assertRating(40.877, 3.840, new_ratings.rating_by_id(1))
self.assertRating(45.493, 2.934, new_ratings.rating_by_id(2))
self.assertRating(19.609, 6.396, new_ratings.rating_by_id(3))
self.assertRating(18.712, 5.625, new_ratings.rating_by_id(4))
self.assertRating(29.353, 7.673, new_ratings.rating_by_id(5))
self.assertRating(9.872, 3.891, new_ratings.rating_by_id(6))
self.assertRating(48.830, 4.590, new_ratings.rating_by_id(7))
self.assertRating(29.813, 1.976, new_ratings.rating_by_id(8))
def test_one_on_seven(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (25.0, 25.0 / 3)
}, {
2: (25.0, 25.0 / 3),
3: (25.0, 25.0 / 3),
4: (25.0, 25.0 / 3),
5: (25.0, 25.0 / 3),
6: (25.0, 25.0 / 3),
7: (25.0, 25.0 / 3),
8: (25.0, 25.0 / 3)
}],
rank=[1, 2])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.000, self.calculator.match_quality(teams, game_info))
self.assertRating(40.582, 7.917, new_ratings.rating_by_id(1))
self.assertRating(9.418, 7.917, new_ratings.rating_by_id(2))
self.assertRating(9.418, 7.917, new_ratings.rating_by_id(3))
self.assertRating(9.418, 7.917, new_ratings.rating_by_id(4))
self.assertRating(9.418, 7.917, new_ratings.rating_by_id(5))
self.assertRating(9.418, 7.917, new_ratings.rating_by_id(6))
self.assertRating(9.418, 7.917, new_ratings.rating_by_id(7))
self.assertRating(9.418, 7.917, new_ratings.rating_by_id(8))
def test_five_teams_of_one_not_drawn(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (25.0, 25.0 / 3)
}, {
2: (25.0, 25.0 / 3)
}, {
3: (25.0, 25.0 / 3)
}, {
4: (25.0, 25.0 / 3)
}, {
5: (25.0, 25.0 / 3)
}],
rank=[1, 2, 3, 4, 5])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.040, self.calculator.match_quality(teams, game_info))
self.assertRating(34.363135705841188, 6.1361528798112692,
new_ratings.rating_by_id(1))
self.assertRating(29.058448805636779, 5.5358352402833413,
new_ratings.rating_by_id(2))
self.assertRating(25.000000000031758, 5.4200805474429847,
new_ratings.rating_by_id(3))
self.assertRating(20.941551194426314, 5.5358352402709672,
new_ratings.rating_by_id(4))
self.assertRating(15.636864294158848, 6.136152879829349,
new_ratings.rating_by_id(5))
def test_eight_teams_of_one_drawn(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (25.0, 25.0 / 3)
}, {
2: (25.0, 25.0 / 3)
}, {
3: (25.0, 25.0 / 3)
}, {
4: (25.0, 25.0 / 3)
}, {
5: (25.0, 25.0 / 3)
}, {
6: (25.0, 25.0 / 3)
}, {
7: (25.0, 25.0 / 3)
}, {
8: (25.0, 25.0 / 3)
}],
rank=[1, 1, 1, 1, 1, 1, 1, 1])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.004, self.calculator.match_quality(teams, game_info))
self.assertRating(25.000, 4.592, new_ratings.rating_by_id(1))
self.assertRating(25.000, 4.592, new_ratings.rating_by_id(2))
self.assertRating(25.000, 4.592, new_ratings.rating_by_id(3))
self.assertRating(25.000, 4.592, new_ratings.rating_by_id(4))
self.assertRating(25.000, 4.592, new_ratings.rating_by_id(5))
self.assertRating(25.000, 4.592, new_ratings.rating_by_id(6))
self.assertRating(25.000, 4.592, new_ratings.rating_by_id(7))
self.assertRating(25.000, 4.592, new_ratings.rating_by_id(8))
def test_four_on_four(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (25.0, 25.0 / 3),
2: (25.0, 25.0 / 3),
3: (25.0, 25.0 / 3),
4: (25.0, 25.0 / 3)
}, {
5: (25.0, 25.0 / 3),
6: (25.0, 25.0 / 3),
7: (25.0, 25.0 / 3),
8: (25.0, 25.0 / 3)
}],
rank=[1, 2])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.447, self.calculator.match_quality(teams, game_info))
self.assertRating(27.198, 8.059, new_ratings.rating_by_id(1))
self.assertRating(27.198, 8.059, new_ratings.rating_by_id(2))
self.assertRating(27.198, 8.059, new_ratings.rating_by_id(3))
self.assertRating(27.198, 8.059, new_ratings.rating_by_id(4))
self.assertRating(22.802, 8.059, new_ratings.rating_by_id(5))
self.assertRating(22.802, 8.059, new_ratings.rating_by_id(6))
self.assertRating(22.802, 8.059, new_ratings.rating_by_id(7))
self.assertRating(22.802, 8.059, new_ratings.rating_by_id(8))
def test_eight_teams_of_one_upset(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (10.0, 8.0)
}, {
2: (15.0, 7.0)
}, {
3: (20.0, 6.0)
}, {
4: (25.0, 5.0)
}, {
5: (30.0, 4.0)
}, {
6: (35.0, 3.0)
}, {
7: (40.0, 2.0)
}, {
8: (45.0, 1.0)
}],
rank=[1, 2, 3, 4, 5, 6, 7, 8])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.000, self.calculator.match_quality(teams, game_info))
self.assertRating(35.135, 4.506, new_ratings.rating_by_id(1))
self.assertRating(32.585, 4.037, new_ratings.rating_by_id(2))
self.assertRating(31.329, 3.756, new_ratings.rating_by_id(3))
self.assertRating(30.984, 3.453, new_ratings.rating_by_id(4))
self.assertRating(31.751, 3.064, new_ratings.rating_by_id(5))
self.assertRating(34.051, 2.541, new_ratings.rating_by_id(6))
self.assertRating(38.263, 1.849, new_ratings.rating_by_id(7))
self.assertRating(44.118, 0.983, new_ratings.rating_by_id(8))
def test_one_on_one(self):
game_info = EloGameInfo(1200, 200)
teams = Match([{1: (1200, 25)}, {2: (1400, 25)}], [1, 2])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.4805, self.calculator.match_quality(teams, game_info))
self.assertRating(1218.99, new_ratings.rating_by_id(1))
self.assertRating(1381.01, new_ratings.rating_by_id(2))
def test_two_player_not_drawn(self):
game_info = TrueSkillGameInfo()
teams = Match([{1: (25.0, 25.0 / 3)}, {2: (25.0, 25.0 / 3)}], [1, 2])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.447214, self.calculator.match_quality(teams, game_info))
self.assertRating(29.39583201999924, 7.171475587326186,
new_ratings.rating_by_id(1))
self.assertRating(20.60416798000076, 7.171475587326186,
new_ratings.rating_by_id(2))
def new_ratings(self, teams, game_info=None):
game_info = TrueSkillGameInfo.ensure_game_info(game_info)
self.validate_team_and_player_counts(teams)
teams.sort()
return Match([
self.new_team_ratings(teams[0], teams[1], teams.comparison(0, 1),
game_info),
self.new_team_ratings(teams[1], teams[0], teams.comparison(1, 0),
game_info)
])
def test_one_on_one_massive_upset_drawn(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (25.0, 25.0 / 3)
}, {
2: (50.0, 25.0 / 2)
}],
rank=[1, 1])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.110, self.calculator.match_quality(teams, game_info))
self.assertRating(31.662, 7.137, new_ratings.rating_by_id(1))
self.assertRating(35.010, 7.910, new_ratings.rating_by_id(2))
def test_two_player_chess_not_drawn(self):
game_info = TrueSkillGameInfo(1200.0, 1200.0 / 3, 200.0, 1200.0 / 300,
0.03)
teams = Match([{
1: (1301.0007, 42.9232)
}, {
2: (1188.7560, 42.5570)
}],
rank=[1, 2])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertRating(1304.7820836053318, 42.843513887848658,
new_ratings.rating_by_id(1))
self.assertRating(1185.0383099003536, 42.485604606897752,
new_ratings.rating_by_id(2))
def test_one_on_two_draw(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (25.0, 25.0 / 3)
}, {
2: (25.0, 25.0 / 3),
3: (25.0, 25.0 / 3)
}],
rank=[1, 1])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.135, self.calculator.match_quality(teams, game_info))
self.assertRating(31.660, 7.138, new_ratings.rating_by_id(1))
self.assertRating(18.340, 7.138, new_ratings.rating_by_id(2))
self.assertRating(18.340, 7.138, new_ratings.rating_by_id(3))
def test_one_on_two_somewhat_balanced(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (40.0, 6.0)
}, {
2: (20.0, 7.0),
3: (25.0, 8.0)
}],
rank=[1, 2])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.478, self.calculator.match_quality(teams, game_info))
self.assertRating(42.744, 5.602, new_ratings.rating_by_id(1))
self.assertRating(16.266, 6.359, new_ratings.rating_by_id(2))
self.assertRating(20.123, 7.028, new_ratings.rating_by_id(3))
def test_two_player_drawn(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (25.0, 25.0 / 3)
}, {
2: (25.0, 25.0 / 3)
}],
rank=[1, 1])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.447, self.calculator.match_quality(teams, game_info))
self.assertRating(25.0, 6.4575196623173081,
new_ratings.rating_by_id(1))
self.assertRating(25.0, 6.4575196623173081,
new_ratings.rating_by_id(2))
def test_three_teams_of_one_drawn(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (25.0, 25.0 / 3)
}, {
2: (25.0, 25.0 / 3)
}, {
3: (25.0, 25.0 / 3)
}],
rank=[1, 1, 1])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.200, self.calculator.match_quality(teams, game_info))
self.assertRating(25.0, 5.698, new_ratings.rating_by_id(1))
self.assertRating(25.0, 5.695, new_ratings.rating_by_id(2))
self.assertRating(25.0, 5.698, new_ratings.rating_by_id(3))
def test_one_on_three_draw(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (25.0, 25.0 / 3)
}, {
2: (25.0, 25.0 / 3),
3: (25.0, 25.0 / 3),
4: (25.0, 25.0 / 3)
}],
rank=[1, 1])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.012, self.calculator.match_quality(teams, game_info))
self.assertRating(34.990, 7.455, new_ratings.rating_by_id(1))
self.assertRating(15.010, 7.455, new_ratings.rating_by_id(2))
self.assertRating(15.010, 7.455, new_ratings.rating_by_id(3))
self.assertRating(15.010, 7.455, new_ratings.rating_by_id(4))
def test_two_on_two(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (25.0, 25.0 / 3),
2: (25.0, 25.0 / 3)
}, {
3: (25.0, 25.0 / 3),
4: (25.0, 25.0 / 3)
}],
rank=[1, 2])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.447, self.calculator.match_quality(teams, game_info))
self.assertRating(28.108, 7.774, new_ratings.rating_by_id(1))
self.assertRating(28.108, 7.774, new_ratings.rating_by_id(2))
self.assertRating(21.892, 7.774, new_ratings.rating_by_id(3))
self.assertRating(21.892, 7.774, new_ratings.rating_by_id(4))
def test_two_on_two_unbalanced_draw(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (15.0, 8.0),
2: (20.0, 6.0)
}, {
3: (25.0, 4.0),
4: (30.0, 3.0)
}],
rank=[1, 1])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.214, self.calculator.match_quality(teams, game_info))
self.assertRating(21.570, 6.556, new_ratings.rating_by_id(1))
self.assertRating(23.696, 5.418, new_ratings.rating_by_id(2))
self.assertRating(23.357, 3.833, new_ratings.rating_by_id(3))
self.assertRating(29.075, 2.931, new_ratings.rating_by_id(4))
def test_two_on_two_upset(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (20.0, 8.0),
2: (25.0, 6.0)
}, {
3: (35.0, 7.0),
4: (40.0, 5.0)
}],
rank=[1, 2])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.084, self.calculator.match_quality(teams, game_info))
self.assertRating(29.698, 7.008, new_ratings.rating_by_id(1))
self.assertRating(30.455, 5.594, new_ratings.rating_by_id(2))
self.assertRating(27.575, 6.346, new_ratings.rating_by_id(3))
self.assertRating(36.211, 4.768, new_ratings.rating_by_id(4))
def new_ratings(self, teams, game_info=None):
game_info = EloGameInfo.ensure_game_info(game_info)
self.validate_team_and_player_counts(teams)
# ensure sorted by rank
teams.sort()
winner, winner_rating = teams[0].player_rating()[0]
loser, loser_rating = teams[1].player_rating()[0]
return Match([Team({winner: self.new_rating(winner_rating,
loser_rating,
teams.comparison(0, 1),
game_info)}),
Team({loser: self.new_rating(loser_rating,
winner_rating,
teams.comparison(1, 0),
game_info)})])
def test_one_on_two_balanced_partial_play(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (25.0, 25.0 / 3)
}, {
(2, 0.5): (25.0, 25.0 / 3),
(3, 0.5): (25.0, 25.0 / 3)
}],
rank=[1, 2])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.4472135955, self.calculator.match_quality(teams, game_info))
self.assertRating(30.1102397877, 6.73461146087,
new_ratings.rating_by_id(1))
self.assertRating(22.4448801061, 7.96412515904,
new_ratings.rating_by_id(2))
self.assertRating(22.4448801061, 7.96412515904,
new_ratings.rating_by_id(3))
def test_three_on_two(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (28.0, 7.0),
2: (27.0, 6.0),
3: (26.0, 5.0)
}, {
4: (30.0, 4.0),
5: (31.0, 3.0)
}],
rank=[1, 2])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.254, self.calculator.match_quality(teams, game_info))
self.assertRating(28.658, 6.770, new_ratings.rating_by_id(1))
self.assertRating(27.484, 5.856, new_ratings.rating_by_id(2))
self.assertRating(26.336, 4.917, new_ratings.rating_by_id(3))
self.assertRating(29.785, 3.958, new_ratings.rating_by_id(4))
self.assertRating(30.879, 2.983, new_ratings.rating_by_id(5))
def test_three_teams_of_one_not_drawn(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (25.0, 25.0 / 3)
}, {
2: (25.0, 25.0 / 3)
}, {
3: (25.0, 25.0 / 3)
}],
rank=[1, 2, 3])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.200, self.calculator.match_quality(teams, game_info))
self.assertRating(31.675352419172107, 6.6559853776206905,
new_ratings.rating_by_id(1))
self.assertRating(25.000000000003912, 6.2078966412243233,
new_ratings.rating_by_id(2))
self.assertRating(18.324647580823971, 6.6559853776218318,
new_ratings.rating_by_id(3))
def test_four_teams_of_one_not_drawn(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (25.0, 25.0 / 3)
}, {
2: (25.0, 25.0 / 3)
}, {
3: (25.0, 25.0 / 3)
}, {
4: (25.0, 25.0 / 3)
}],
rank=[1, 2, 3, 4])
new_ratings = self.calculator.new_ratings(teams, game_info)
self.assertMatchQuality(
0.089, self.calculator.match_quality(teams, game_info))
self.assertRating(33.206680965631264, 6.3481091698077057,
new_ratings.rating_by_id(1))
self.assertRating(27.401454693843323, 5.7871629348447584,
new_ratings.rating_by_id(2))
self.assertRating(22.598545306188374, 5.7871629348413451,
new_ratings.rating_by_id(3))
self.assertRating(16.793319034361271, 6.3481091698144967,
new_ratings.rating_by_id(4))
def new_ratings(self, matches, rating_period=None, game_info=None):
game_info = GlickoGameInfo.ensure_game_info(game_info)
# get unique list of players and ensure ratings are consistant
players = {}
opponents = defaultdict(list)
player_rd = {}
# Step 1: calculate r and RD for onset of rating period
# also cache values of g(RDj) and E(s|r, rj, RDj) for each player
for match in matches:
# ensure winning team is team 0
match.sort()
for player, rating in match.player_rating():
if player in players:
if rating != players[player]:
raise ValueError(
"Inconsistant ratings: player %s has rating %s and rating %s"
% (player, rating, players[player]))
else:
# calc new RD from old RD
if (self.c_factor is None
or rating.last_rating_period is None
or rating_period is None):
player_rd[player] = rating.stdev
else:
t = rating_period - rating.last_rating_period
if t <= 0:
raise ValueError(
"Player %s has a last_rating_period equal to the current rating_period"
% (player))
player_rd[player] = min(
game_info.beta,
sqrt(rating.stdev**2 + self.c_factor * t))
players[player] = rating
# create opponent lists of players and outcomes
player1 = match[0].players()[0]
player2 = match[1].players()[0]
opponents[player1].append(
(player2, GlickoCalculator.score[match.comparison(0, 1)]))
opponents[player2].append(
(player1, GlickoCalculator.score[match.comparison(1, 0)]))
# Step 2: carry out the update calculations for each player separately
q = log(10.0) / 400.0
def g(rd):
return 1.0 / sqrt(1.0 + 3.0 * q**2.0 * (rd**2.0) / pi**2.0)
def E(r, rj, rd_j):
return 1.0 / (1.0 + pow(10.0, -g(rd_j) * (r - rj) / 400.0))
def d2(g_rd, e_sr_r_rd):
return pow(
q**2.0 * sum(g_rd[j]**2.0 * e_sr_r_rd[j] * (1.0 - e_sr_r_rd[j])
for j in range(len(g_rd))), -1.0)
new_ratings = Match()
for player, rating in players.items():
# cache values of g(RDj) and E(s|r, r, RDj) for each opponent
opponent_g_rd = []
opponent_e_sr_r_rd = []
opponent_s = []
for opponent, score in opponents[player]:
opponent_g_rd.append(g(player_rd[opponent]))
opponent_e_sr_r_rd.append(
E(rating.mean, players[opponent].mean,
player_rd[opponent]))
opponent_s.append(score)
# cache value, this form used twice in the paper
rd2_d2 = (1.0 / player_rd[player]**2.0 +
1.0 / d2(opponent_g_rd, opponent_e_sr_r_rd))
# new rating value
r_new = (rating.mean +
q / rd2_d2 * sum(opponent_g_rd[j] *
(opponent_s[j] - opponent_e_sr_r_rd[j])
for j in range(len(opponent_s))))
# new rating deviation value
rd_new = sqrt(pow(rd2_d2, -1))
new_ratings.append(
Team({player: GlickoRating(r_new, rd_new, rating_period)}))
return new_ratings
def test_sixteen_teams_of_one_not_drawn(self):
game_info = TrueSkillGameInfo()
teams = Match([{
1: (25.0, 25.0 / 3)
}, {
2: (25.0, 25.0 / 3)
}, {
3: (25.0, 25.0 / 3)
}, {
4: (25.0, 25.0 / 3)
}, {
5: (25.0, 25.0 / 3)
}, {
6: (25.0, 25.0 / 3)
}, {
7: (25.0, 25.0 / 3)
}, {
8: (25.0, 25.0 / 3)
}, {
9: (25.0, 25.0 / 3)
}, {
10: (25.0, 25.0 / 3)
}, {
11: (25.0, 25.0 / 3)
}, {
12: (25.0, 25.0 / 3)
}, {
13: (25.0, 25.0 / 3)
}, {
14: (25.0, 25.0 / 3)
}, {
15: (25.0, 25.0 / 3)
}, {
16: (25.0, 25.0 / 3)
}],
rank=list(range(16)))
new_ratings = self.calculator.new_ratings(teams, game_info)
# naive matrix implementation too slow with 16 teams
if not hasattr(Matrix, 'is_square'):
self.assertMatchQuality(
0.000, self.calculator.match_quality(teams, game_info))
self.assertRating(40.53945776946920, 5.27581643889050,
new_ratings.rating_by_id(1))
self.assertRating(36.80951229454210, 4.71121217610266,
new_ratings.rating_by_id(2))
self.assertRating(34.34726355544460, 4.52440328139991,
new_ratings.rating_by_id(3))
self.assertRating(32.33614722608720, 4.43258628279632,
new_ratings.rating_by_id(4))
self.assertRating(30.55048814671730, 4.38010805034365,
new_ratings.rating_by_id(5))
self.assertRating(28.89277312234790, 4.34859291776483,
new_ratings.rating_by_id(6))
self.assertRating(27.30952161972210, 4.33037679041216,
new_ratings.rating_by_id(7))
self.assertRating(25.76571046519540, 4.32197078088701,
new_ratings.rating_by_id(8))
self.assertRating(24.23428953480470, 4.32197078088703,
new_ratings.rating_by_id(9))
self.assertRating(22.69047838027800, 4.33037679041219,
new_ratings.rating_by_id(10))
self.assertRating(21.10722687765220, 4.34859291776488,
new_ratings.rating_by_id(11))
self.assertRating(19.44951185328290, 4.38010805034375,
new_ratings.rating_by_id(12))
self.assertRating(17.66385277391300, 4.43258628279643,
new_ratings.rating_by_id(13))
self.assertRating(15.65273644455550, 4.52440328139996,
new_ratings.rating_by_id(14))
self.assertRating(13.19048770545810, 4.71121217610273,
new_ratings.rating_by_id(15))
self.assertRating(9.46054223053080, 5.27581643889032,
new_ratings.rating_by_id(16))
def new_ratings(self, matches, rating_period=None, game_info=None):
game_info = GlickoGameInfo.ensure_game_info(game_info)
# get unique list of players and ensure ratings are consistant
players = {}
opponents = defaultdict(list)
player_rd = {}
# Step 1: calculate r and RD for onset of rating period
# also cache values of g(RDj) and E(s|r, rj, RDj) for each player
for match in matches:
# ensure winning team is team 0
match.sort()
for player, rating in match.player_rating():
if player in players:
if rating != players[player]:
raise ValueError("Inconsistant ratings: player %s has rating %s and rating %s" % (player, rating, players[player]))
else:
# calc new RD from old RD
if (self.c_factor is None or
rating.last_rating_period is None or
rating_period is None):
player_rd[player] = rating.stdev
else:
t = rating_period - rating.last_rating_period
if t <= 0:
raise ValueError("Player %s has a last_rating_period equal to the current rating_period" % (player))
player_rd[player] = min(game_info.beta, sqrt(rating.stdev ** 2 + self.c_factor * t))
players[player] = rating
# create opponent lists of players and outcomes
player1 = match[0].players()[0]
player2 = match[1].players()[0]
opponents[player1].append((player2, GlickoCalculator.score[match.comparison(0, 1)]))
opponents[player2].append((player1, GlickoCalculator.score[match.comparison(1, 0)]))
# Step 2: carry out the update calculations for each player separately
q = log(10.0) / 400.0
def g(rd):
return 1.0 / sqrt(1.0 + 3.0 * q ** 2.0 * (rd ** 2.0) / pi ** 2.0)
def E(r, rj, rd_j):
return 1.0 / (1.0 + pow(10.0, -g(rd_j) * (r - rj) / 400.0))
def d2(g_rd, e_sr_r_rd):
return pow(q ** 2.0 * sum(
g_rd[j] ** 2.0 * e_sr_r_rd[j] * (1.0 - e_sr_r_rd[j])
for j in range(len(g_rd))
), -1.0)
new_ratings = Match()
for player, rating in players.items():
# cache values of g(RDj) and E(s|r, r, RDj) for each opponent
opponent_g_rd = []
opponent_e_sr_r_rd = []
opponent_s = []
for opponent, score in opponents[player]:
opponent_g_rd.append(g(player_rd[opponent]))
opponent_e_sr_r_rd.append(E(rating.mean, players[opponent].mean, player_rd[opponent]))
opponent_s.append(score)
# cache value, this form used twice in the paper
rd2_d2 = (1.0 / player_rd[player] ** 2.0 +
1.0 / d2(opponent_g_rd, opponent_e_sr_r_rd))
# new rating value
r_new = (rating.mean + q / rd2_d2 * sum(
opponent_g_rd[j] * (opponent_s[j] - opponent_e_sr_r_rd[j])
for j in range(len(opponent_s))
))
# new rating deviation value
rd_new = sqrt(pow(rd2_d2, -1))
new_ratings.append(Team({player: GlickoRating(r_new, rd_new, rating_period)}))
return new_ratings