metastan: Compile metastan code into JAGS code

In mjskay/metabayes: Meta-Languages for Specifying Bayesian Models (JAGS, Stan) as R Code

Description

Compiles metastan code (specified as an R expression) into JAGS code.

Usage

metastan(...)

Arguments

...	Metastan code for named Stan blocks, specified as bare R expressions. The name of the argument to this function is used as the name of the Stan block in the compiled code, with underscores replaced with spaces (e.g. transformed_parameters becomes the block transformed parameters).

Details

metastan integrates Stan model specification more easily into R by allowing Stan models to be specified as bare R code rather than as character strings. Because R and Stan are syntactically similar, with only a few exceptions, metastan models look very similar to Stan models. The primary exception is variable declaration, where the Stan syntax is C-like (type variable), and looks like this:

int a; real<lower=0> x; vector<lower=-1,upper=1>[3,3] c[10];

By contrast, metastan reverses the declaration order (variable : type), for example:

a : int x : real(lower=0) c[10] : vector(lower=-1,upper=1)[3,3]

This syntax has the advantage that the order you read subscripts in is the same as the order they are declared in (e.g., in the above example, c has subscripts c[10,3,3], not c[3,3,10]).

Compared to using strings to specify Stan code in R, this approach approach has the advantage that syntax checking in R editors helps prevent simple errors without having to attempt to compile the model with Stan, decreasing turnaround time when iterating on models.

Value

An object of class c("metastan", "metamodel"). Metastan models have the following functions:

code	Returns the JAGS code for this model as a character string.

Author(s)

Matthew Kay

See Also

See also code for extracting the resulting Stan model as a character string, and metajags for the JAGS equivalent of this function.

Examples

## Not run: 
library(rstan)

# FIRST, using metastan:
model = metastan(
    data = {
        N : int(lower=0)
        x : vector[N]
        y : vector[N]
    },
    parameters = {
        alpha : real
        beta : real
        sigma : real(lower=0)
    },
    model = {
        y ~ normal(alpha + beta * x, sigma)
    }
)

#(some code setting up data_list, etc) 
#...

stan_fit = stan(code(model), data=data_list, ...)


#SECOND, the traditional way: The above metastan approach 
#is equivalent to the following more traditional approach:

model_string = "
    data {
        int<lower=0> N;
        vector[N] x;
        vector[N] y;
    }
    parameters {
        real alpha;
        real beta;
        real<lower=0> sigma;
    }
    model {
        y ~ normal(alpha + beta * x, sigma);
    }
"

stan_fit = stan(model_string, data=data_list, ...)

## End(Not run)

mjskay/metabayes documentation built on May 23, 2019, 1:05 a.m.