Assign sequence to integer array

Modeling
1

Minute syntax question. Let’ say I want to create a sequence of integers. The following doesn’t work:

int sequence[n] = 1:n;

and the error message is the following

SYNTAX ERROR, MESSAGE(S) FROM PARSER:
Variable definition base type mismatch, variable declared as base type int[ ] variable definition has base type intVariable definition dimensions mismatch, definition specifies 1, declaration specifies 0 error in 'model873b2e593919_monster_reparm_par' at line 215, column 34
  -------------------------------------------------
   213:   real C_inh[N_exh];
   214:   
   215:   int exp_index[N_experiments] = 1:N_experiments;  // for reduce_sum
                                         ^
   216: 
  -------------------------------------------------

Error in stanc(filename, allow_undefined = TRUE) : 
  failed to parse Stan model 'monster_reparm_par' due to the above error.

It’s easy to do this in a for loop, but I was wondering if the above syntax is expected to work.

2

I don’t think it is. Unless something has changed in the latest stan versions, Stan doesn’t use range-expressions like R does. E.g., R takes 1:10 and expands it to c(1,2,3,4,...,10). Stan does not do this. It’s really just some syntactic sugar. E.g., the for(n in 1:N) does not expand out to 1,2,…,10; but rather something like for(int n = 1; n <= N; ++n); so A:B is really just another way of setting a starting value and a condition.

Does that make sense?

3
4

That’s something I’ve wanted to do before, and would make for a nice convenience function.

I can add this to the Math library pretty easily, @rok_cesnovar would this be difficult for the parser? In other words, would it be difficult to deduce the different meanings of 1:N between:

int seq[N] = 1:N;

and

for(i in 1:N)
1 Like
5

As of 2.24 there are some utility construction utilities like

real[] linspaced_array(int, real, real)
row_vector linspaced_row_vector(int, real, real)
vector linspaced_vector(int, real, real)

All arguments here are only allowed to be data.

linspaced_array(K, low, high) produces an array from low to high (included) with elements spaced as (high - low) / (K - 1). For K=1, the array will contain the high value, for K=0 it returns an empty array.

These were added both in Math (PR) and stanc3 (PR) but is unfortunately missing docs: https://github.com/stan-dev/docs/issues/206

We might be able to parse 1:N to linspaced_*(N,1,N). I am however not an expert on parsing to say that with much authority. Please open a stanc3 issue if you would prefer 1:N to linspaced_array.

1 Like
6

Yeah that’s what I was thinking. It just needs to be tweaked because the current implementation will only return an array of doubles (i.e., std::vector<double> / real[]) which can’t be used for indexing or assigned to an integer array.

But that just needs some simple templating added to make that work

7

Ohh yeah, skimmed this to fast. If int is important then yes, we would need to make int[] linspaced_int_array(int, int, int), or int[] linspaced_array(int, int, int)

8

The colon is allowed in both for-loops and index expressions. In both cases an integer array is also a valid expressions but the ambiguity does not change behaviour; either interpretation does the same thing.

vector[N] svec = vec[1:N];
for (i in 1:N) {

int seq[N] = (1:N);
vector[N] svec = vec[seq];
for (i in seq) {

linspaced does not quite work with integers. For example, linspaced_array(3,1,2) should be {1.0,1.5,2.0} but those aren’t integers.
Instead you want something like int[] unitspaced_array(int low, int high).

1 Like
9

Oh is this already available?:

int seq[N] = (1:N);
10

Doh, of course. Silly me. Good call

11

No. but we could make it available and remove the other uses of : and no one would notice.