restore previous model

code/models/blocks/confusion_model_binomial_hurdle_fast.stan

@@ -7,8 +7,7 @@ real confusion_model_lpdf(array[] matrix lambda,
     array[] int group,
     array[] real age,
     array[] real clip_duration,
-    matrix p,
-    matrix etas_p
+    matrix p
 ) {
     real ll = 0;
     vector [4] dp;
@@ -36,19 +35,19 @@ real confusion_model_lpdf(array[] matrix lambda,
                     // (as opposed to two vocs)
                     for (chi1 in max(0, 2*chi_d-algo[k,i]):chi_d) {
                         int chi = 2*chi_d-chi1;
-                        bp[1] = chi_d==0?0:beta_binomial_lpmf(chi1 | chi_d, p[1,i]*etas_p[1,i], (1-p[1,i])*etas_p[1,i]);
+                        bp[1] = chi_d==0?0:beta_lpmf(chi1 | chi_d, p[1,i]);
 
                         for (och_d in 0:(truth[k,2]>0?min(truth[k,2], algo[k,i]-chi):0)) {
                             dp[2] = truth[k,2]==0?0:binomial_lpmf(och_d | truth[k,2], lambda[lg,2,i]);
                             for (och1 in max(0, 2*och_d-(algo[k,i]-chi)):och_d) {
                                 int och = 2*och_d-och1;
-                                bp[2] = och_d==0?0:beta_binomial_lpmf(och1 | och_d, p[2,i]*etas_p[2,i], (1-p[2,i])*etas_p[2,i]);
+                                bp[2] = och_d==0?0:beta_lpmf(och1 | och_d, p[2,i]);
 
                                 for (fem_d in 0:(truth[k,3]>0?min(truth[k,3], algo[k,i]-chi-och):0)) {
                                     dp[3] = truth[k,3]==0?0:binomial_lpmf(fem_d | truth[k,3], lambda[lg,3,i]);
                                     for (fem1 in max(0, 2*fem_d-(algo[k,i]-chi-och)):fem_d) {
                                         int fem = 2*fem_d-fem1;
-                                        bp[3] = fem_d==0?0:beta_binomial_lpmf(fem1 | fem_d, p[3,i]*etas_p[3,i], (1-p[3,i])*etas_p[3,i]);
+                                        bp[3] = fem_d==0?0:beta_lpmf(fem1 | fem_d, p[3,i]);
 
                                         for (mal_d in 0:(truth[k,4]>0?min(truth[k,4], algo[k,i]-chi-och-fem):0)) {
                                             dp[4] = truth[k,4]==0?0:binomial_lpmf(mal_d | truth[k,4], lambda[lg,4,i]);
@@ -56,7 +55,7 @@ real confusion_model_lpdf(array[] matrix lambda,
                                                 int mal = 2*mal_d-mal1;
                                                 int delta = algo[k,i] - (mal+fem+och+chi);
                                                 if (delta==0) {
-                                                    bp[4] = mal_d==0?0:beta_binomial_lpmf(mal1 | mal_d, p[4,i]*etas_p[4,i], (1-p[4,i])*etas_p[4,i]);
+                                                    bp[4] = mal_d==0?0:beta_lpmf(mal1 | mal_d, p[4,i]);
                                                     log_contrib_comb[n] += sum(dp+bp);
                                                     n = n+1;
                                                 }

code/models/blocks/confusion_model_parameters_binomial_hurdle.stan

@@ -1,5 +1,4 @@
-matrix<lower=10>[n_classes,n_classes] etas;
-matrix<lower=10>[n_classes,n_classes] etas_p;
+matrix<lower=1>[n_classes,n_classes] etas;
 matrix<lower=0,upper=1>[n_classes,n_classes] mus;
 array [n_groups] matrix<lower=0,upper=1>[n_classes,n_classes] lambda;
 matrix<lower=0,upper=1>[n_classes,n_classes] p;

code/models/blocks/confusion_model_priors_binomial_hurdle.stan

@@ -2,7 +2,6 @@ for (i in 1:n_classes) {
     p[i] ~ beta(3,1);
     mus[i] ~ uniform(0, 1);
     etas[i] ~ pareto(1, 1.5);
-    etas_p[i] ~ pareto(1, 1.5);
 }
 
 for (c in 1:n_groups) {
@@ -11,5 +10,5 @@ for (c in 1:n_groups) {
         ll_normal += beta_proportion_lpdf(lambda[c,i,:] | mus[i,:], etas[i,:]);
     }
     // tolerance to outliers
-    target += log_mix(0.025, 0, ll_normal);
+    target += log_mix(0.02, 0, ll_normal);
 }

code/models/dev_siblings_binomial_hurdle_fast.stan

@@ -116,7 +116,7 @@ model {
         confusion_model_lpdf, lambda, 1,
         n_classes, n_clips,
         algo_total, truth_total, group, clip_duration, clip_age,
-        p, etas_p//, lambda_fp
+        p//, lambda_fp
     );
 
     // priors on the nuisance parameters of the confusion model