correct approximate EKF-MCMC

Author: helske

src/mgg_ssm.cpp

@@ -472,6 +472,7 @@ arma::cube mgg_ssm::simulate_states() {
     um(j) = normal(engine);
   }
   asim.slice(0).col(0) = L_P1 * um;
+  arma::vec y_tmp = y;
   for (unsigned int t = 0; t < (n - 1); t++) {
     arma::uvec na_y = arma::find_nonfinite(y.col(t));
     if (na_y.n_elem < p) {
@@ -500,6 +501,6 @@ arma::cube mgg_ssm::simulate_states() {
   }
 
   asim.slice(0) += fast_smoother();
-  
+  y = y_tmp;
   return asim;
 }

src/nlg_amcmc.cpp

@@ -431,8 +431,7 @@ void nlg_amcmc::gaussian_sampling(nlg_ssm model, const unsigned int n_threads) {
     gaussian_state_sampler(model, theta_storage, mode_storage,
       alpha_storage);
   }
-  posterior_storage = prior_storage + approx_loglik_storage - scales_storage + 
-    log(weight_storage);
+  posterior_storage = prior_storage + approx_loglik_storage;
 }
 
 void nlg_amcmc::gaussian_state_sampler(nlg_ssm model,
@@ -477,8 +476,7 @@ void nlg_amcmc::gaussian_state_sampler(nlg_ssm model,
     }
     approx_model.compute_HH();
     approx_model.compute_RR();
-    
-    alpha.slice(i) = approx_model.simulate_states();
+    alpha.slice(i) = approx_model.simulate_states().slice(0).t();
   }
 }
 

vignettes/growth_model.Rmd

@@ -118,7 +118,6 @@ model <- nlg_ssm(y = y, a1=pntrs$a1, P1 = pntrs$P1,
 
 Let's first run Extended Kalman filter and smoother using our initial guess for $\theta$:
 ```{r ekf}
-library(diagis)
 out_filter <- ekf(model)
 out_smoother <- ekf_smoother(model)
 ts.plot(cbind(y, out_filter$att[, 2], out_smoother$alphahat[, 2]), col = 1:3)
@@ -127,12 +126,13 @@ ts.plot(cbind(out_filter$att[, 1], out_smoother$alphahat[, 1]), col = 1:2)
 
 ## Markov chain Monte Carlo
 
-For parameter inference, we can perform full Bayesian inference with \texttt{bssm}. There are multiple choices for the MCMC algorithm in the package, and here we will use $\psi$-PF based pseudo-marginal MCMC with delayed acceptance. Let us compare this approach with EKF-based approximate MCMC:
+For parameter inference, we can perform full Bayesian inference with \texttt{bssm}. There are multiple choices for the MCMC algorithm in the package, and here we will use $\psi$-PF based pseudo-marginal MCMC with delayed acceptance. Let us compare this approach with EKF-based approximate MCMC (note that the number of MCMC iterations is rather small):
 
 ```{r mcmc}
-out_mcmc_pm <- run_mcmc(model, n_iter = 1e4, nsim_states = 10, method = "pm", 
+out_mcmc_pm <- run_mcmc(model, n_iter = 5000, nsim_states = 10, method = "pm", 
   simulation_method = "psi")
-out_mcmc_ekf <- run_mcmc(model, n_iter = 1e4, method = "ekf")
+out_mcmc_ekf <- run_mcmc(model, n_iter = 5000, method = "ekf")
+library("diagis")
 alpha_ekf <- weighted_mean(out_mcmc_ekf$alpha, out_mcmc_ekf$counts)
 alpha_pm <- weighted_mean(out_mcmc_pm$alpha, out_mcmc_pm$counts)
 ts.plot(cbind(y, alpha_pm[, 2], alpha_ekf[, 2]), col = 1:3)