6.11 Higher-order functions
There are several expression constructions in Stan that act as higher-order functions.4
The higher-order functions and the signature of their argument functions are listed in the higher-order functions table and the variadic higher-order functions table.
Higher-order Functions Table. Higher-order functions in Stan with their argument function types. The first group of arguments can be a function of parameters or data. The second group of arguments, consisting of a real and integer array in all cases, must be expressions involving only data and literals.
function | parameter or data args | data args | return type |
---|---|---|---|
algebra_solver |
vector, vector |
real[], int[] |
vector |
algebra_solver_newton |
vector, vector |
real[], int[] |
vector |
integrate_1d , |
real, real, real[] |
real[], int[] |
real |
integrate_ode_X , |
real, real[], real[] |
real[], int[] |
real[] |
map_rect |
vector, vector |
real[], int[] |
vector |
For example, the integrate_ode_rk45
function can be used to integrate
differential equations in Stan:
functions {
real[] foo(real t, real[] y, real[] theta, real[] x_r, int[] x_i) {
...
...
int<lower=1> T;
real y0[2];
real t0;
real ts[T];
real theta[1];
real x_r[0];
int x_i[0];
...
real y_hat[T, 2] = integrate_ode_rk45(foo, y0, t0, ts, theta, x_r, x_i);
The function argument is foo
, the name of the user-defined
function; as shown in the higher-order functions table,
integrate_ode_rk45
takes a real array, a real, three more real arrays, and
an integer array as arguments and returns 2D real array.
Variadic Higher-order Functions Table. Variadic Higher-order functions in Stan with their argument function types. The first group of arguments are restricted in type. The sequence of trailing arguments can be of any length with any types.
function | restricted args | return type |
---|---|---|
ode_X , |
vector, real, real[] |
vector[] |
reduce_sum |
T[], T1, T2 |
real |
T
, T1
, and T2
can be any Stan type.
For example, the ode_rk45
function can be used to integrate
differential equations in Stan:
functions {
vector foo(real t, vector y, real theta, vector beta, int[] x_i, int index) {
...
...
int<lower=1> T;
vector[2] y0;
real t0;
real ts[T];
real theta;
vector[7] beta;
int x_i[10];
int index;
...
vector[2] y_hat[T] = ode_rk45(foo, y0, t0, ts, theta, beta, x_i, index);
The function argument is foo
, the name of the user-defined
function. As shown in the
variadic higher-order functions table,
ode_rk45
takes a real, a vector, a real, a real array, and a sequence of
arguments whos types match those at the end of foo
and returns an array of
vectors.
Functions passed by reference
The function argument to higher-order functions is always passed as the first argument. This function argument must be provided as the name of a user-defined or built-in function. No quotes are necessary.
Data-restricted arguments
Some of the arguments to higher-order functions are restricted to data. This means they must be expressions containing only data variables, transformed data variables, or literals; the may contain arbitrary functions applied to data variables or literals, but must not contain parameters, transformed parameters, or local variables from any block other than transformed data.
For user-defined functions the qualifier data
may be prepended
to the type to restrict the argument to data-only variables.
Internally, they are implemented as their own expression types because Stan doesn’t have object-level functional types (yet).↩