e2c.py 文件源码

def cov(self):
        """This should only be called when NormalDistribution represents one sample"""
        if self.v is not None and self.r is not None:
            assert self.v.dim() == 1
            dim = self.v.dim()
            v = self.v.unsqueeze(1)  # D * 1 vector
            rt = self.r.unsqueeze(0)  # 1 * D vector
            A = torch.eye(dim) + v.mm(rt)
            return A.mm(torch.diag(self.sigma.pow(2)).mm(A.t()))
        else:
            return torch.diag(self.sigma.pow(2))