llm.py 文件源码

def simulation_smoother(self,beta):
        """ Koopman's simulation smoother - simulates from states given
        model latent variables and observations

        Parameters
        ----------

        beta : np.array
            Contains untransformed starting values for latent variables

        Returns
        ----------
        - A simulated state evolution
        """         

        T, Z, R, Q, H = self._ss_matrices(beta)

        # Generate e_t+ and n_t+
        rnd_h = np.random.normal(0,np.sqrt(H),self.data.shape[0]+1)
        q_dist = ss.multivariate_normal([0.0], Q)
        rnd_q = q_dist.rvs(self.data.shape[0]+1)

        # Generate a_t+ and y_t+
        a_plus = np.zeros((T.shape[0],self.data.shape[0]+1)) 
        a_plus[0,0] = np.mean(self.data[0:5])
        y_plus = np.zeros(self.data.shape[0])

        for t in range(0,self.data.shape[0]+1):
            if t == 0:
                a_plus[:,t] = np.dot(T,a_plus[:,t]) + rnd_q[t]
                y_plus[t] = np.dot(Z,a_plus[:,t]) + rnd_h[t]
            else:
                if t != self.data.shape[0]:
                    a_plus[:,t] = np.dot(T,a_plus[:,t-1]) + rnd_q[t]
                    y_plus[t] = np.dot(Z,a_plus[:,t]) + rnd_h[t]

        alpha_hat,_ = self.smoothed_state(self.data,beta)
        alpha_hat_plus,_ = self.smoothed_state(y_plus,beta)
        alpha_tilde = alpha_hat - alpha_hat_plus + a_plus

        return alpha_tilde