base_models.py 文件源码

def predict_forward_mm(self, T, x_control):
        """Summary

        Args:
            T (TYPE): Description
            x_control (None, optional): Description

        Returns:
            TYPE: Description
        """
        mx = np.zeros((T, self.Din))
        vx = np.zeros((T, self.Din))
        my = np.zeros((T, self.Dout))
        vy_noiseless = np.zeros((T, self.Dout))
        vy = np.zeros((T, self.Dout))
        post_m, post_v = self.get_posterior_x()
        mtm1 = post_m[[-1], :]
        vtm1 = post_v[[-1], :]
        for t in range(T):
            if self.Dcon_dyn > 0:
                mtm1 = np.hstack((mtm1, x_control[[t], :]))
                vtm1 = np.hstack((vtm1, np.zeros((1, self.Dcon_dyn))))
            mt, vt = self.dyn_layer.forward_prop_thru_post(mtm1, vtm1)
            if self.Dcon_emi > 0:
                mtc = np.hstack((mt, x_control[[t], :]))
                vtc = np.hstack((vt, np.zeros((1, self.Dcon_emi))))
            else:
                mtc, vtc = mt, vt
            if self.gp_emi:
                mft, vft = self.emi_layer.forward_prop_thru_post(mtc, vtc)
                myt, vyt_n = self.lik_layer.output_probabilistic(mft, vft)
            else:
                myt, vyt, vyt_n = self.emi_layer.output_probabilistic(mt, vt)
                vft = np.diagonal(vyt, axis1=1, axis2=2)
                vyt_n = np.diagonal(vyt_n, axis1=1, axis2=2)
            mx[t, :], vx[t, :] = mt, vt
            my[t, :], vy_noiseless[t, :], vy[t, :] = myt, vft, vyt_n
            mtm1 = mt
            vtm1 = vt
        return mx, vx, my, vy_noiseless, vy