def generatePoisson(self, team1: str, team2: str, league: str, show: bool = False):
team1Ratings, team2Ratings, avgGoals, avgGA = self.getRatings(team1, team2, league)
# Calculate expected goals for each team
team1xG: float = team1Ratings[0] * team2Ratings[1] * avgGoals
team2xG: float = team2Ratings[0] * team1Ratings[1] * avgGoals
# Create 2d array for discrete probabilities, with team1 as columns and team2 as rows
distribution: np.ndarray((6, 6)) = np.zeros((6, 6))
for i in range(6):
for j in range(6):
team1p = poisson.pmf(j, team1xG)
team2p = poisson.pmf(i, team2xG)
if j == 5:
team1p = 1 - poisson.cdf(5, team1xG)
if i == 5:
team2p = 1 - poisson.cdf(5, team2xG)
distribution[i][j] = team1p * team2p
if show:
plt.imshow(distribution)
plt.show()
return (distribution, team1xG, team2xG)
def generatePoisson(self, show: bool = False) -> np.ndarray:
try:
self.team1Ratings
self.team2Ratings
except AttributeError:
self.getRatings()
# Calculate expected goals for each team
team1xG: float = self.team1Ratings[0] * \
self.team2Ratings[1] * self.avgGoals
team2xG: float = self.team2Ratings[0] * \
self.team1Ratings[1] * self.avgGoals
# Create 2d array for discrete probabilities, with team1 as columns and team2 as rows
distribution: np.ndarray((6, 6)) = np.zeros((6, 6))
for i in range(6):
for j in range(6):
team1p = poisson.pmf(j, team1xG)
team2p = poisson.pmf(i, team2xG)
if j == 5:
team1p = 1 - poisson.cdf(5, team1xG)
if i == 5:
team2p = 1 - poisson.cdf(5, team2xG)
distribution[i][j] = team1p * team2p
if show:
plt.imshow(distribution)
plt.show()
self.dist = distribution
self.team1xG = team1xG
self.team2xG = team2xG
return distribution
def independentBivPois2(A, B, alpha=0.01):
robjects.r["set.seed"](1337)
m1 = numpy.array(A)
m2 = numpy.array(B)
bivpois = importr('bivpois')
mu1 = mean(m1)
mu2 = mean(m2)
numpy2ri.activate()
# print(bivpois.pbivpois(robjects.r["as.vector"](m1), robjects.r["as.vector"](m2), robjects.r["c"](mu1,mu2,0.00000001)))
bp = bivpois.simple_bp(robjects.r["as.vector"](m1),
robjects.r["as.vector"](m2),
maxit=35)
dpll = max(bp[1])
idll = numpy.sum(numpy.log(poisson.pmf(m1, mu=mu1))) + numpy.sum(
numpy.log(poisson.pmf(m2, mu=mu2)))
lrt = -2.0 * (idll - dpll)
chisq = chi2.ppf(1.0 - alpha, 1)
print(bp[0])
print(dpll)
print(idll)
print(lrt)
print(chisq)
# is independent
return lrt <= chisq
def getOneWayPoisson(self, league: str, team1: str, team2: str) -> Union[list, list]:
dist, team1xG, team2xG = self.generatePoisson(team1, team2, league)
team1Poisson = []
team2Poisson = []
for i in range(5):
team1Poisson.append(poisson.pmf(i, team1xG))
team2Poisson.append(poisson.pmf(i, team2xG))
team1Poisson.append(1-poisson.cdf(4, team1xG))
team2Poisson.append(1-poisson.cdf(4, team2xG))
return (team1Poisson, team2Poisson)
def getOneWayPoisson(self) -> Union[list, list]:
try:
self.distribution
except AttributeError:
self.generatePoisson()
team1Poisson = []
team2Poisson = []
for i in range(5):
team1Poisson.append(poisson.pmf(i, self.team1xG))
team2Poisson.append(poisson.pmf(i, self.team2xG))
team1Poisson.append(1-poisson.cdf(4, self.team1xG))
team2Poisson.append(1-poisson.cdf(4, self.team2xG))
return [team1Poisson, team2Poisson]
def getLogLikelihood(self, documents, debug=False):
ll = 0
if documents.ndim > 1:
for i in range(0, documents.shape[0]):
ll += self.getLogLikelihood(documents[i])
else:
# it's only one
document = documents
lambdas = self.getLambdas(document)
if debug:
print(document)
print(numpy.round(lambdas, 0))
for j in range(0, self.nF):
pmfval = poisson.pmf(document[j], lambdas[j])
if pmfval == 0.0:
pmfval = numpy.finfo(float).eps
ll += math.log(pmfval)
return ll