Negative binomial won't recognize int dispersion parameter

Modeling
1

For some reason Stan doesn’t like that I’ve given assigned the value 1 to the phi argument in the negative binomial distribution. I find this perplexing since other packages like brms use this exact syntax when you specify a geometric distribution for the response data.

model {

  // priors including all constants
  for(s in 1:S){
    for(t in s:S){
      target += beta_lpdf(P[s,t] | 0.5, 0.5) ;
    }
  }
  // likelihood including all constants
  if(!prior_only){
     target += neg_binomial_2_log_lpmf(Y | eta, 1) ;
  }
}

SYNTAX ERROR, MESSAGE(S) FROM PARSER:
No matches for:

neg_binomial_2_log_lpmf(matrix, matrix, int)

Available argument signatures for neg_binomial_2_log_lpmf:

neg_binomial_2_log_lpmf(int, real, real)
neg_binomial_2_log_lpmf(int, real, real[ ])
neg_binomial_2_log_lpmf(int, real, vector)
neg_binomial_2_log_lpmf(int, real, row_vector)
neg_binomial_2_log_lpmf(int, real[ ], real)
neg_binomial_2_log_lpmf(int, real[ ], real[ ])
neg_binomial_2_log_lpmf(int, real[ ], vector)
neg_binomial_2_log_lpmf(int, real[ ], row_vector)
neg_binomial_2_log_lpmf(int, vector, real)
neg_binomial_2_log_lpmf(int, vector, real[ ])
neg_binomial_2_log_lpmf(int, vector, vector)
neg_binomial_2_log_lpmf(int, vector, row_vector)
neg_binomial_2_log_lpmf(int, row_vector, real)
neg_binomial_2_log_lpmf(int, row_vector, real[ ])
neg_binomial_2_log_lpmf(int, row_vector, vector)
neg_binomial_2_log_lpmf(int, row_vector, row_vector)
neg_binomial_2_log_lpmf(int[ ], real, real)
neg_binomial_2_log_lpmf(int[ ], real, real[ ])
neg_binomial_2_log_lpmf(int[ ], real, vector)
neg_binomial_2_log_lpmf(int[ ], real, row_vector)
neg_binomial_2_log_lpmf(int[ ], real[ ], real)
neg_binomial_2_log_lpmf(int[ ], real[ ], real[ ])
neg_binomial_2_log_lpmf(int[ ], real[ ], vector)
neg_binomial_2_log_lpmf(int[ ], real[ ], row_vector)
neg_binomial_2_log_lpmf(int[ ], vector, real)
neg_binomial_2_log_lpmf(int[ ], vector, real[ ])
neg_binomial_2_log_lpmf(int[ ], vector, vector)
neg_binomial_2_log_lpmf(int[ ], vector, row_vector)
neg_binomial_2_log_lpmf(int[ ], row_vector, real)
neg_binomial_2_log_lpmf(int[ ], row_vector, real[ ])
neg_binomial_2_log_lpmf(int[ ], row_vector, vector)
neg_binomial_2_log_lpmf(int[ ], row_vector, row_vector)

error in ‘model428418796aa8_Orchid_Stan_Code’ at line 40, column 49

38:   // likelihood including all constants
39:   if (!prior_only) {
40:     target += neg_binomial_2_log_lpmf(Y | eta, 1) ;
                                                    ^
41:   }

Error in stanc(file = file, model_code = model_code, model_name = model_name, :
failed to parse Stan model ‘Orchid Stan Code’ due to the above error.

2

Yeah, just do 1.0. What you wrote should be legal, but it is easy enough to work around.

3

I’ve already tried that. It didn’t fix the issue.

SYNTAX ERROR, MESSAGE(S) FROM PARSER:
No matches for:

neg_binomial_2_log_lpmf(matrix, matrix, real)

Available argument signatures for neg_binomial_2_log_lpmf:

neg_binomial_2_log_lpmf(int, real, real)
neg_binomial_2_log_lpmf(int, real, real[ ])
neg_binomial_2_log_lpmf(int, real, vector)
neg_binomial_2_log_lpmf(int, real, row_vector)
neg_binomial_2_log_lpmf(int, real[ ], real)
neg_binomial_2_log_lpmf(int, real[ ], real[ ])
neg_binomial_2_log_lpmf(int, real[ ], vector)
neg_binomial_2_log_lpmf(int, real[ ], row_vector)
neg_binomial_2_log_lpmf(int, vector, real)
neg_binomial_2_log_lpmf(int, vector, real[ ])
neg_binomial_2_log_lpmf(int, vector, vector)
neg_binomial_2_log_lpmf(int, vector, row_vector)
neg_binomial_2_log_lpmf(int, row_vector, real)
neg_binomial_2_log_lpmf(int, row_vector, real[ ])
neg_binomial_2_log_lpmf(int, row_vector, vector)
neg_binomial_2_log_lpmf(int, row_vector, row_vector)
neg_binomial_2_log_lpmf(int[ ], real, real)
neg_binomial_2_log_lpmf(int[ ], real, real[ ])
neg_binomial_2_log_lpmf(int[ ], real, vector)
neg_binomial_2_log_lpmf(int[ ], real, row_vector)
neg_binomial_2_log_lpmf(int[ ], real[ ], real)
neg_binomial_2_log_lpmf(int[ ], real[ ], real[ ])
neg_binomial_2_log_lpmf(int[ ], real[ ], vector)
neg_binomial_2_log_lpmf(int[ ], real[ ], row_vector)
neg_binomial_2_log_lpmf(int[ ], vector, real)
neg_binomial_2_log_lpmf(int[ ], vector, real[ ])
neg_binomial_2_log_lpmf(int[ ], vector, vector)
neg_binomial_2_log_lpmf(int[ ], vector, row_vector)
neg_binomial_2_log_lpmf(int[ ], row_vector, real)
neg_binomial_2_log_lpmf(int[ ], row_vector, real[ ])
neg_binomial_2_log_lpmf(int[ ], row_vector, vector)
neg_binomial_2_log_lpmf(int[ ], row_vector, row_vector)

error in ‘model428474bc19f9_Orchid_Stan_Code’ at line 39, column 49

37:   }
38:   // likelihood including all constants
39:   target += neg_binomial_2_log_lpmf(Y | eta, 1.0) ;
                                                    ^
40:

Error in stanc(file = file, model_code = model_code, model_name = model_name, :
failed to parse Stan model ‘Orchid Stan Code’ due to the above error.

4

Now it does not like the matrices (although the int may have been ok before). Your Y needs to be a two-dimensional int array rather than a matrix. And you may need to loop over the first dimension, so it would be something like

for (j in 1:J) target += neg_binomial_2_log_lpmf(Y[j] | eta[ , j] 1);
5

Great. This does fix the issue. Though, it did open up other issues in the code I had to address before I could be sure.

Related, how does one multiple an int with a real number? This issue came up when fixing my code. The solution I came up with was to simply keep a real valued duplicate of the int data, but this is not an elegant solution.

1 Like
6

int * real should definitely work. What was the line of code in question and parser error?

7

Below is the error I received. Everything works fine when a create a duplicate copy of Y that is a real matrix instead of a 2d int array.

SYNTAX ERROR, MESSAGE(S) FROM PARSER:
No matches for:

dot_product(int[ ], row_vector)

Available argument signatures for dot_product:

dot_product(vector, vector)
dot_product(row_vector, row_vector)
dot_product(vector, row_vector)
dot_product(row_vector, vector)
dot_product(real[ ], real[ ])

error in ‘modelf1838707a2c_Orchid_Stan_Code’ at line 24, column 79

22:   for(n in 1:N){
23:     for(s in 1:S){
24:         logpcond[n,s] = log(P[s,s]) + dot_product(Y[n,(s+1):],log(P[s,(s+1):])) ;
                                                                                  ^
25:     }

Error in stanc(file = file, model_code = model_code, model_name = model_name, :
failed to parse Stan model ‘Orchid Stan Code’ due to the above error.

1 Like
8

int * real should work but int[] * vector won’t. It will if you do

dot_product(to_vector(Y[n, (s + 1):]), log(P[s,(s+1):]));
1 Like
9

Great. It works now. Thanks for your help.