def run(self, data, verbose=False):
'''
Override function from superclass.
:param data: Pandas dataframe on which to test model.
:return: void
'''
r = self.player_rankings
p = self.params
glicko = rs.Glicko2(p["mu"], p["phi"], p["sigma"], p["tau"], p["eps"])
probabilities = []
bet_amount = []
P = []
Q = []
for i, row in data.iterrows():
name1 = row["Winner"]
name2 = row["Loser"]
wsp1 = row["WSP1"]
wsp2 = row["WSP2"]
# If there is no data on serve win percentage, skip.
if math.isnan(wsp1) or math.isnan(wsp2):
probabilities.append(0.5)
bet_amount.append(0)
continue
if not r.get(name1, False):
r[name1] = [glicko.create_rating(), glicko.create_rating()]
if not r.get(name2, False):
r[name2] = [glicko.create_rating(), glicko.create_rating()]
p = glicko.expect(r[name1][0], r[name2][1])
q = glicko.expect(r[name2][0], r[name1][1])
P.append(p)
Q.append(q)
win_prob = calc.prob_win_match(p, q, row["Best_of"])
probabilities.append(win_prob)
bet_amount.append(1)
r[name1][0], r[name2][1] = glicko.rate_1vs1(r[name1][0], r[name2][1], wsp1 )
r[name2][0], r[name1][1] = glicko.rate_1vs1(r[name2][0], r[name1][1], wsp2)
self.player_rankings = r
if verbose:
sez = sorted([[rt[0].mu, rt[1].mu, key] for key, rt in r.iteritems()], key=lambda x: x[0]+x[1], reverse=True)
print '\n'.join(['{:>22s} {:8.1f} {:8.1f}'.format(p[2], p[0], p[1]) for p in sez[:20]]) + "\n\n"
df = data.copy()
df["win_prob"] = probabilities
df["bet_amount"] = bet_amount
return df
def run(self, data, verbose=False):
'''
Override function from superclass.
:param data: Pandas dataframe on which to test model.
:return: void
'''
r = self.player_rankings
probabilities = []
bet_amount = []
for i, row in data.iterrows():
name1 = row["Winner"]
name2 = row["Loser"]
wsp1 = row["WSP1"]
wsp2 = row["WSP2"]
# If there is no data on serve win percentage, skip.
if math.isnan(wsp1) or math.isnan(wsp2):
probabilities.append(0.5)
bet_amount.append(0)
continue
if not r.get(name1, False):
r[name1] = Player()
if not r.get(name2, False):
r[name2] = Player()
s1, r1 = r[name1].get()
s2, r2 = r[name2].get()
p = s1 + r2 - 0.4
q = s2 + r1 - 0.4
win_prob = calc.prob_win_match(p, q, row["Best_of"])
probabilities.append(win_prob)
bet_amount.append(1)
r[name1].add(wsp1, 1.0 - wsp2)
r[name2].add(wsp2, 1.0 - wsp1)
self.player_rankings = r
if verbose:
sez = sorted([[rt.get()[0], rt.get()[1], key] for key, rt in r.iteritems()], key=lambda x: x[0]+x[1], reverse=True)
print '\n'.join(['{:>22s} {:8.5f} {:8.5f}'.format(p[2], p[0], p[1]) for p in sez[:20]]) + "\n\n"
df = data.copy()
df["win_prob"] = probabilities
df["bet_amount"] = bet_amount
return df
def win_probabilities(self, player1, player2, best_of=3):
'''
Calculates fair betting odds according to our model and returns them
as tuple.
:param player1: Class player for player1
:param player2: Class player for player2
:param best_of: How many sets we are playing
:return: Tuple (odds1, odds2), representing fair odds for player
one or two winning
'''
p, q = self.expect(player1, player2)
win_prob = calc.prob_win_match(p, q, best_of)
return (win_prob, 1 - win_prob, p, q)
def run(self, data, verbose=False):
'''
Override function from superclass.
:param data: Pandas dataframe on which to test model.
:return: void
'''
r = self.player_rankings
p = self.params
glicko = rs.Glicko2(p["mu"], p["phi"], p["sigma"], p["tau"], p["eps"])
probabilities = []
bet_amount = []
P = []
Q = []
for i, row in data.iterrows():
name1 = row["Winner"]
name2 = row["Loser"]
wsp1 = row["WSP1"]
wsp2 = row["WSP2"]
# If there is no data on serve win percentage, skip.
if math.isnan(wsp1) or math.isnan(wsp2):
probabilities.append(0.5)
bet_amount.append(0)
continue
if not r.get(name1, False):
r[name1] = [glicko.create_rating(), glicko.create_rating()]
if not r.get(name2, False):
r[name2] = [glicko.create_rating(), glicko.create_rating()]
p = glicko.expect(r[name1][0], r[name2][1])
q = glicko.expect(r[name2][0], r[name1][1])
P.append(p)
Q.append(q)
win_prob = calc.prob_win_match(p, q, row["Best_of"])
probabilities.append(win_prob)
bet_amount.append(1)
r[name1][0], r[name2][1] = glicko.rate_1vs1(
r[name1][0], r[name2][1], wsp1)
r[name2][0], r[name1][1] = glicko.rate_1vs1(
r[name2][0], r[name1][1], wsp2)
self.player_rankings = r
if verbose:
sez = sorted([[rt[0].mu, rt[1].mu, key]
for key, rt in r.iteritems()],
key=lambda x: x[0] + x[1],
reverse=True)
print '\n'.join([
'{:>22s} {:8.1f} {:8.1f}'.format(p[2], p[0], p[1])
for p in sez[:20]
]) + "\n\n"
df = data.copy()
df["win_prob"] = probabilities
df["bet_amount"] = bet_amount
return df
qg = game() p = p_t confirm = calculations._prob_win_game(p) print "p:", p print_table(memo1) print "Confirming with calculations: ", confirm print # Tiebreaker. pt = tiebreaker() confirm = calculations._prob_win_tiebreaker(p, q) print "pt:", pt print_table(memo2) print "Confirming with calculations: ", confirm print # Set. ps = set() confirm = calculations._prob_win_set(p, q) print "pg, qg:", pg, qg print_table(memo3) print "Confirming with calculations: ", confirm print # Match. pm = match(best_of=5) confirm = calculations.prob_win_match(p, q, 5) print "pm, ", pm print_table(memo4) print "Confirming with calculations: ", confirm print
def run(self, data, verbose=False):
'''
Override function from superclass.
:param data: Pandas dataframe on which to test model.
:return: void
'''
r = self.player_rankings
g = self.player_games
elo = rs.Elo(self.params["mu"], self.params["K"])
edge = self.params["edge"]
probabilities = []
bet_amount = []
for i, row in data.iterrows():
name1 = row["Winner"]
name2 = row["Loser"]
wsp1 = row["WSP1"]
wsp2 = row["WSP2"]
# If there is no data on serve win percentage, skip.
if math.isnan(wsp1) or math.isnan(wsp2):
probabilities.append(0.5)
bet_amount.append(0)
continue
if not r.get(name1, False):
r[name1] = [elo.create_rating(), elo.create_rating()]
if not r.get(name2, False):
r[name2] = [elo.create_rating(), elo.create_rating()]
p = elo.expect(r[name1][0], r[name2][1]) + edge
q = elo.expect(r[name2][0], r[name1][1]) + edge
win_prob = calc.prob_win_match(p, q, row["Best_of"])
probabilities.append(win_prob)
# probabilities.append(random.random())
# bet_amount.append(1)
g[name1] = g.get(name1, 0.0) + 1
g[name2] = g.get(name2, 0.0) + 1
if g[name1] > 50 and g[name2] > 50:
bet_amount.append(1)
else:
bet_amount.append(0)
r[name1][0], r[name2][1] = elo.match(r[name1][0], r[name2][1],
wsp1 - edge)
r[name2][0], r[name1][1] = elo.match(r[name2][0], r[name1][1],
wsp2 - edge)
self.player_rankings = r
self.player_games = g
if verbose:
sez = sorted([[rt[0].rating, rt[1].rating, key]
for key, rt in r.iteritems()],
key=lambda x: x[0] + x[1],
reverse=True)
print '\n'.join([
'{:>22s} {:8.1f} {:8.1f}'.format(str(p[2]), p[0], p[1])
for p in sez[:20]
]) + "\n\n"
df = data.copy()
df["win_prob"] = probabilities
df["bet_amount"] = bet_amount
return df
def run(self, data, verbose=False):
'''
Override function from superclass.
:param data: Pandas dataframe on which to test model.
:return: void
'''
r = self.player_rankings
s = self.surface_advantage
g = self.player_games
elo = rs.Elo(self.params["mu"], self.params["K"])
edge = self.params["edge"]
probabilities = []
bet_amount = []
for i, row in data.iterrows():
name1 = row["Winner"]
name2 = row["Loser"]
wsp1 = row["WSP1"]
wsp2 = row["WSP2"]
# If there is no data on serve win percentage, skip.
if math.isnan(wsp1) or math.isnan(wsp2):
probabilities.append(0.5)
bet_amount.append(0)
continue
if not r.get(name1, False):
r[name1] = [elo.create_rating(), elo.create_rating()]
s[name1] = Player()
if not r.get(name2, False):
r[name2] = [elo.create_rating(), elo.create_rating()]
s[name2] = Player()
s1, r1, h11, h12 = s[name1].get_advantage(row["Surface"])
s2, r2, h21, h22 = s[name2].get_advantage(row["Surface"])
h1 = h11 + h22
h2 = h12 + h21
a1 = s1 - r2
if a1 + h1 < 0.0: a1 += h1
elif a1 - h1 > 0.0: a1 -= h1
else: a1 = 0
a2 = s2 - r1
if a2 + h2 < 0.0: a2 += h2
elif a2 - h2 > 0.0: a2 -= h2
else: a2 = 0
p = elo.expect(r[name1][0], r[name2][1]) + edge + a1
q = elo.expect(r[name2][0], r[name1][1]) + edge + a2
s[name1].update(wsp1, 1.0 - wsp2, row["Surface"])
s[name2].update(wsp2, 1.0 - wsp1, row["Surface"])
win_prob = calc.prob_win_match(p, q, row["Best_of"])
g[name1] = g.get(name1, 0.0) + 1
g[name2] = g.get(name2, 0.0) + 1
"""
if g[name1]>5 and g[name2]>5:
bet_amount.append(1)
else:
bet_amount.append(0)
"""
bet_amount.append(1)
probabilities.append(win_prob)
r[name1][0], r[name2][1] = elo.match(r[name1][0], r[name2][1], wsp1 - edge)
r[name2][0], r[name1][1] = elo.match(r[name2][0], r[name1][1], wsp2 - edge)
self.surface_advantage = s
self.player_rankings = r
self.player_games = g
if verbose:
sez = sorted([[rt[0].rating, rt[1].rating, str(key)] for key, rt in r.iteritems()], key=lambda x: x[0]+x[1], reverse=True)
print '\n'.join(['{:>22s} {:8.1f} {:8.1f}'.format(p[2], p[0], p[1]) for p in sez[:20]]) + "\n\n"
df = data.copy()
df["win_prob"] = probabilities
df["bet_amount"] = bet_amount
return df
pg = game() p_t = p; p = q; qg = game(); p = p_t confirm = calculations._prob_win_game(p) print "p:", p print_table(memo1) print "Confirming with calculations: ", confirm print # Tiebreaker. pt = tiebreaker() confirm = calculations._prob_win_tiebreaker(p, q) print "pt:", pt print_table(memo2) print "Confirming with calculations: ", confirm print # Set. ps = set() confirm = calculations._prob_win_set(p, q) print "pg, qg:", pg, qg print_table(memo3) print "Confirming with calculations: ", confirm print # Match. pm = match(best_of=5) confirm = calculations.prob_win_match(p, q, 5) print "pm, ", pm print_table(memo4) print "Confirming with calculations: ", confirm print
def run(self, data, verbose=False):
"""
Override function from superclass.
:param data: Pandas dataframe on which to test model.
:return: void
"""
r = self.player_rankings
g = self.player_games
elo = rs.Elo(self.params["mu"], self.params["K"])
edge = self.params["edge"]
probabilities = []
bet_amount = []
for i, row in data.iterrows():
name1 = row["Winner"]
name2 = row["Loser"]
wsp1 = row["WSP1"]
wsp2 = row["WSP2"]
# If there is no data on serve win percentage, skip.
if math.isnan(wsp1) or math.isnan(wsp2):
probabilities.append(0.5)
bet_amount.append(0)
continue
if not r.get(name1, False):
r[name1] = [elo.create_rating(), elo.create_rating()]
if not r.get(name2, False):
r[name2] = [elo.create_rating(), elo.create_rating()]
p = elo.expect(r[name1][0], r[name2][1]) + edge
q = elo.expect(r[name2][0], r[name1][1]) + edge
win_prob = calc.prob_win_match(p, q, row["Best_of"])
probabilities.append(win_prob)
# probabilities.append(random.random())
# bet_amount.append(1)
g[name1] = g.get(name1, 0.0) + 1
g[name2] = g.get(name2, 0.0) + 1
if g[name1] > 50 and g[name2] > 50:
bet_amount.append(1)
else:
bet_amount.append(0)
r[name1][0], r[name2][1] = elo.match(r[name1][0], r[name2][1], wsp1 - edge)
r[name2][0], r[name1][1] = elo.match(r[name2][0], r[name1][1], wsp2 - edge)
self.player_rankings = r
self.player_games = g
if verbose:
sez = sorted(
[[rt[0].rating, rt[1].rating, key] for key, rt in r.iteritems()],
key=lambda x: x[0] + x[1],
reverse=True,
)
print "\n".join(["{:>22s} {:8.1f} {:8.1f}".format(str(p[2]), p[0], p[1]) for p in sez[:20]]) + "\n\n"
df = data.copy()
df["win_prob"] = probabilities
df["bet_amount"] = bet_amount
return df
__author__ = 'riko' import time import calculations as calc p = 0.70 q = 0.70 start = time.clock() m1 = calc.prob_win_match(p, q) m2 = calc.monte_carlo_match(p, q) end = time.clock() print m1 print m2 print "Time: ", end - start
