Expose_stan_functions C++ code not compliling

Modeling
1

I am trying to use integrate_ode_rkf to perform a double definite integration \int_{0}^{x}\int_{0}^{y}f(x, y)dydx. The basic idea is that the first integration wrt y takes place in a user-defined function (of x) and that function is then integrated wrt x to give the result.
The following Stan code is apparently ‘syntactically correct’:

functions {

 
// Example integrand
  real[] integrand_ode(real y, real[] f, real[] theta, real[] x_r, int[] x_i) {
    real df_dx[1]; 
    df_dx[1] = exp(-square(y) - x_r[1]^2) ;
    return(df_dx);
  }
//User-defined function of x after y integrated out
  real ode_integrate(real x, real[] f, real[] theta, real[] xx_r, int[] x_i) {
     
    return(integrate_ode_rk45(integrand_ode, 
                            rep_array(0.0, 1), 
                            0, 
                            rep_array(1.0, 1), 
                            rep_array(0.0, 0), 
                            rep_array(x, 1), 
                            x_i)[1,1]);
  }
  // Final integration wrt x
  real ode_integrate2( ) {
    return(integrate_ode_rk45( ode_integrate, 
                            rep_array(0.0, 1),
                            0, 
                            rep_array(1.0, 1), 
                            rep_array(0.0, 0),
                            rep_array(0.0, 0) , 
                            rep_array(0, 0))[1,1]);

  }
}
data {
}
model {}

but when I use expose_stan_functions an error is generated. In the masses of output (which I can happily attach if it is needed) the key points seem to be

 In file included from /Users/Jeremy1/Rlibs/StanHeaders/include/stan/math/prim/arr.hpp:37:
/Users/Jeremy1/Rlibs/StanHeaders/include/stan/math/prim/arr/functor/coupled_ode_system.hpp:98:15: error: no viable overloaded '='
        dy_dt = f_(t, y, theta_dbl_, x_, x_int_, msgs_);
        ~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/Users/Jeremy1/Rlibs/BH/include/boost/numeric/odeint/stepper/dense_output_runge_kutta.hpp:329:13: note: in instantiation of member function 'stan::math::coupled_ode_system<ode_integrate_functor__, double, double>::operator()' requested here
            sys( get_current_state() , get_current_deriv() , m_t );
            ^
 
                                                                                     ^
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/include/c++/v1/vector:537:13: note: candidate function not viable: no known conversion from 'double' to 'const std::__1::vector<double, std::__1::allocator<double> >' for 1st argument
    vector& operator=(const vector& __x);

The problem is with the final stage of the integration but I do not know what the error messages mean.

2

This seems to be a bug. I’ll look into it.

3

I think there is a bug in expose_stan_functions where it does not display error messages when the Stan program is malformed (@Krzysztof_Sakrejda, @martinmodrak) but if you try to parse it (under rstan 2.18.1 at least), you get

SYNTAX ERROR, MESSAGE(S) FROM PARSER:

first argument to integrate_ode_rk45 must be the name of a function with signature (real, real[], real[], real[], int[]) : real[]   error in 'model512a615354b7_integral' at line 29, column 44
  -------------------------------------------------
    27:                             rep_array(0.0, 0),
    28:                             rep_array(0.0, 0) , 
    29:                             rep_array(0, 0))[1,1]);
                                                   ^
    30: 
  -------------------------------------------------

PARSER EXPECTED: ")"

Look in the user manual for the correct signatures for such functions, but post again if you can’t get it figured out.

4

Thanks, Ben. I’ll look into that.
My code parses without comment. I am using rstan 2,17.4. Does that mean I should update something?

5

Now updated to 2.18.1 and I get the same parser error.

6

I now have code that works:

functions {

 
// Example integrand
  real[] integrand_ode(real y, real[] f, real[] theta, real[] x, int[] x_i) {
    real df_dx[1]; // Do not change
    df_dx[1] = exp(-y^2 - x[1]^2); // Function of x and y to be integrated
    return(df_dx);
  }
//User-defined function of x after y integrated out
  real[] first_integral(real x, real[] f, real[] theta, real[] x_r, int[] x_i) {     
    return(integrate_ode_rk45(integrand_ode, 
                            rep_array(0.0, 1), // Do not change
                            0.0, // ymin
                            rep_array(1.0, 1), //(ymax, 1)
                            rep_array(0.0, 0), // Do not change
                            rep_array(x, 1), // Do not change
                            x_i)[1,]);// Do not change
  }
  // Final integration wrt x
  real second_integral(real xmin, real xmax ) {
    int x_i[0];
    return(integrate_ode_rk45(  first_integral, 
                            rep_array(0.0, 1),// Do not change
                            xmin,   //xmin
                            rep_array(xmax, 1), // (xmax, 1)
                            rep_array(0.0, 0),// Do not change
                            rep_array(0.0, 0) , // Do not change
                            x_i)[1,1]);// Do not change
  }
}
data {
}
model {}
7

I’ll check on it.