The following variables have undefined values

Modeling
1

I am getting the following message when fitting my Stan model (an ecological metapopulation dynamics model) that suggests there is an issue with an object tmp_proj in my generated quantities block:

The following variables have undefined values:  tmp_proj[1,2,2],The following variables have undefined values:  tmp_proj[2,2,2],The following variables have undefined values:  tmp_proj[3,2,2],The following variables have undefined values:  tmp_proj[4,2,2]

… And so on. I am filling in this object with a lot of nested conditional statements, and I suspect I just made a mistake somewhere, but for the life of me I can’t find it! Here’s the part of the generated quantities block where I create it (I can post the whole block, but it’s pretty long). Can anyone spot a mistake? It’s defined as real tmp_proj[np,n_ages, ny_proj];. Based on the message above, I think the issue starts after the if(a==1){} condition (because the first issue happens for tmp_proj[1,2,2] which is patch 1, age 2, and year 2).

d, sigma_r, mean_recruits, rec_dev_proj, T_adjust_proj, proj_init, and z are all estimated or defined earlier. np, ny_proj, and n_ages are the dimensions.

      for(p in 1:np){
        
        for(y in 1:ny_proj){
          
          if(y==1){
            // initiate projection with fixed observation
            for(a in 1:n_ages){
              tmp_proj[p,a, 1] = proj_init[p,a];
            }
          } else { // add case for all other years
          
          for(a in 1:n_ages){
            
            // project age 1
            if(a==1){
              tmp_proj[p,1, y] = mean_recruits * exp(rec_dev_proj[y-1] - pow(sigma_r,2)/2) * T_adjust_proj[p,y]; 
// these values are all estimated or defined earlier, and I know this part is working at least
            }
            // project non-reproductive ages
            else if(a < age_at_maturity){
              tmp_proj[p,a,y] = tmp_proj[p,a-1,y-1] * (1-z);// these just grow and die in the patch
            }
            // project reproductive ages
            else{
              if(p==1){
                tmp_proj[p,a,y] = tmp_proj[p, a-1, y-1] * (1-z) * (1-d) + tmp_proj[p+1, a-1, y-1] * (1-z) * d;
              } // close patch 1 case 
              
              else if(p==np){
                tmp_proj[p,a,y] = tmp_proj[p, a-1, y-1] * (1-z) * (1-d) + tmp_proj[p-1, a-1, y-1] * (1-z) * d;
              } // close highest patch
              
              else{
                tmp_proj[p,a,y] = tmp_proj[p, a-1, y-1] * (1-z) * 2*(1-d) + tmp_proj[p-1, a-1, y-1] * (1-z) * d + tmp_proj[p+1, a-1, y-1] * (1-z) * d;
                
              } // close if/else for main patches
              
            } // close else for reproductive age group
            
            
          } // close age loop
          } // close else for all years except the initial one 
        } // close year loop
      } // close patch loop
2

On mobile so I’m having trouble parsing, but a tip to check is to pre-fill all the elements of tmp_proj in a series of simple nested loops first with a dummy value you’ll recognize as different from real values (ex. 99999999) then put a print at the end of the GQ block to see what elements get left at that value.

1 Like
3

My collaborator figured it out – I just needed to flip the order of the p and the y loops at the top so y was outside.