clustering.py 文件源码

def _EM_VMM_GMM_maximisation_step(self):
        '''Maximisation step SH : 7June2013
        '''
        self.pi_hat = np.sum(self.zij,axis=0)/float(self.n_instances)
        self.mu_list_old = self.mu_list.copy()
        self.kappa_list_old = self.kappa_list.copy()
        self.mean_list_old = self.mean_list.copy()
        self.std_list_old = self.std_list.copy()

        for cluster_ident in range(self.n_clusters):
            inst_tmp = (self.instance_array_complex.T * self.zij[:,cluster_ident]).T
            N= np.sum(self.zij[:,cluster_ident])
            #calculate the best fit for this cluster - all dimensions at once.... using new approximations
            #VMM part
            self.kappa_list[cluster_ident,:], self.mu_list[cluster_ident,:], scale_fit1 = EM_VMM_calc_best_fit(inst_tmp, lookup=self.bessel_lookup_table, N=N)
            #GMM part
            self.std_list[cluster_ident,:], self.mean_list[cluster_ident,:] = EM_GMM_calc_best_fit(self.instance_array_amps, self.zij[:,cluster_ident])
        #Prevent ridiculous situations happening....
        self.kappa_list = np.clip(self.kappa_list,0.1,300)
        self.std_list = np.clip(self.std_list,0.001,300)
        self._EM_VMM_GMM_check_convergence()