def expectation(self, data, model):
"""Compute probabilities
store in self.model
"""
mus = model.mu
sigma = model.sigma
weight = model.weight
probabilities = []
rho = np.zeros(len(data)) # rho is probabilities
for r in range(len(data)):
row = data[r]
x_less = np.array(row) - np.array(mus)
# the expectation <Zim> * P(x|Zim)
rho[r] = multivariate_normal(sigma, x_less, .001) * weight
#rho[r] = scipy.stats.multivariate_normal.pdf(data[r], mean=mus, cov=sigma) * weight
#model.probabilities = hw3.univariate_normal(data, np.std(data), model.mu, model.weight)
print 'RHO'
print rho
# log likelihood
# llh = np.log(rho[i]/sum(rho))
sum_rho = sum(rho)
new_rho = []
#return [float(rho[r])/sum_rho * weight for r in range(len(rho))]
for r in range(len(rho)):
new_rho.append(rho[r]/sum_rho) # * weight)
return new_rho
def multivariate_normal(self, covar_matrix, x_less, alpha=1, mus=[]):
"""
:param d: number of rows in X
:param covar_matrix:
:param x_less: X - u , u is a vector of mu
:return:
"""
return multivariate_normal(covar_matrix, x_less, alpha=1)
