def update_u_v(self, rho):
D = self.D
zeta = self.prior['zeta']
# compute u, v
self.u = self.prior['u'] + (D / 2) * (
1 + np.sum(rho, 0)) + self.zeta * self.k / D * d_hyp1f1(
0.5, D / 2, self.zeta * self.k, iteration=self.max_hy1f1_iter)
self.v = self.prior['v'] + np.sum(rho, 0) * (
D / (2 * self.k) + (1 / D) * d_hyp1f1(0.5, D / 2, self.k, iteration=self.max_hy1f1_iter)) + \
(D / (2 * self.k) + (zeta / D) * d_hyp1f1(0.5, D / 2, zeta * self.k, iteration=self.max_hy1f1_iter))
def update_u_v(self):
# compute u, v
D = self.D
zeta = self.prior['zeta']
self.u = self.prior['u'] + (D / 2) * (
1 + self.temp_k_ss) + self.zeta * self.k * (d_hyp1f1(
0.5, D / 2, self.zeta * self.k, iteration=self.max_hy1f1_iter)
/ D)
self.v = self.prior['v'] + self.temp_k_ss * (
D / (2 * self.k) + (1 / D) * d_hyp1f1(0.5, D / 2, self.k, iteration=self.max_hy1f1_iter)) + \
(D / (2 * self.k) + (zeta / D) * d_hyp1f1(0.5, D / 2, zeta * self.k, iteration=self.max_hy1f1_iter))
def caclulate_log_lik_x(self, x):
D = self.D
E_k = digamma(self.u) - np.log(self.v)
kdk1 = d_hyp1f1(0.5,
D / 2,
self.zeta * self.k,
iteration=self.max_hy1f1_iter)
kdk2 = d_hyp1f1(1.5, (D + 2) / 2,
self.zeta * self.k,
iteration=self.max_hy1f1_iter) * kdk1
kdk3 = d_hyp1f1(0.5, D / 2, self.k, iteration=self.max_hy1f1_iter)
temp = (1 / D * kdk1 + self.zeta * self.k *
(3 / ((D + 2) * D) * kdk2 -
(1 / (D**2)) * kdk1 * kdk1)) * self.k * (E_k + np.log(
self.zeta) - np.log(self.prior['zeta'] * self.k))
log_lik_x = gammaln(
D / 2) - (D / 2) * np.log(2 * np.pi) + (D / 2) * E_k - np.log(
(self.k**(D / 2)) * hyp1f1(0.5, D / 2, self.k)
) - (D / 2 / self.k + 1 / D * kdk3) * (
self.u / self.v - self.k) + self.k / D * kdk1 + temp * (x.dot(
self.xi.T)**2)
return log_lik_x