model.py 文件源码

def update_V(m_opts, m_vars):
    P, N = E_x_omega_col(m_opts, m_vars) # expectation of xi_{nl} for n = i, expecation of omega_{nl} for n = i
    Kappa = PG_col(m_opts, m_vars) # polyagamma kappa_{nl} for n = i
    PN = P*N
    PK = P*Kappa

    for i in range(m_vars['n_labels']):
        PN_i = PN[i][:,np.newaxis]
        PK_i = PK[i]

        sigma = m_vars['U_batch'].T.dot(PN_i*m_vars['U_batch'])# + m_opts['lam_v']*np.eye(m_opts['n_components'])
        x = m_vars['U_batch'].T.dot(PK_i)

        m_vars['sigma_v'][i] = (1-m_vars['gamma'])*m_vars['sigma_v'][i] + m_vars['gamma']*sigma
        m_vars['x_v'][i] = (1-m_vars['gamma'])*m_vars['x_v'][i] + m_vars['gamma']*x
        m_vars['V'][i] = linalg.solve(m_vars['sigma_v'][i], m_vars['x_v'][i])