models.py 文件源码

def _vlb(self):
        vlb = 0.
        vlb += sum(l.get_vlb() for l in self.labels_list)
        vlb += self.weights.get_vlb()
        vlb += sum(c.get_vlb() for c in self.components)
        for l in self.labels_list:
            vlb += np.sum([r.dot(c.expected_log_likelihood(l.data))
                                for c,r in zip(self.components, l.r.T)])

        # add in symmetry factor (if we're actually symmetric)
        if len(set(type(c) for c in self.components)) == 1:
            vlb += special.gammaln(len(self.components)+1)

        return vlb

    ### SVI