expose_functions from brms throws $ operator error

user_16322999 0
Viewed 79

I was working on some code assembled by a colleague and ran into some problems with the expose_functions() function in brms. When I got the error (once I had made sure I hadn’t fat-fingered something), I started a new R session, made sure all of my packages, etc. were updated, and then tried it again - same error.

At this point, I went to the vignette (here) and ran that code - where the error repeated itself exactly.

Here are the relevant parts of the vignette:

# Data
data("cbpp", package = "lme4")

# Custom family definition
beta_binomial2 <- custom_family(
        "beta_binomial2", dpars = c("mu", "phi"),
        links = c("logit", "log"), lb = c(NA, 0),
        type = "int", vars = "vint1[n]")

# Stan function definition
stan_funs <- "
        real beta_binomial2_lpmf(int y, real mu, real phi, int T) {
                return beta_binomial_lpmf(y | T, mu * phi, (1 - mu) * phi);
        }
        int beta_binomial2_rng(real mu, real phi, int T) {
                return beta_binomial_rng(T, mu * phi, (1 - mu) * phi);
        }
        "

# Stan variables
stanvars <- stanvar(scode = stan_funs, block = "functions")

# Model fit
fit2 <- brm(
        incidence | vint(size) ~ period + (1|herd), 
        data = cbpp, 
        family = beta_binomial2, 
        stanvars = stanvars
)

# Expose functions
expose_functions(fit2, vectorize = TRUE)

None of this is original code - it’s taken directly from the vignette. When I run this, I generate 32 warnings and one error:

Error: $ operator is invalid for atomic vectors

This is exactly the same thing that happens when I run my other code. I’m sure it’s me and not the code, but I don’t see where I’ve gone wrong here.

Just in case it helps anyone find the solution, here is a list of the various warnings (unique warnings only). Most of them seem "harmless" (based on this), but some may not be.

In file included from file656664715164.cpp:8:
In file included from /Library/Frameworks/R.framework/Versions/4.1/Resources/library/RcppEigen/include/RcppEigen.h:25:
In file included from /Library/Frameworks/R.framework/Versions/4.1/Resources/library/RcppEigen/include/RcppEigenForward.h:40:
In file included from /Library/Frameworks/R.framework/Versions/4.1/Resources/library/RcppEigen/include/unsupported/Eigen/SparseExtra:51:
/Library/Frameworks/R.framework/Versions/4.1/Resources/library/RcppEigen/include/unsupported/Eigen/../../Eigen/src/Core/util/ReenableStupidWarnings.h:14:30: warning: pragma diagnostic pop could not pop, no matching push [-Wunknown-pragmas]
    #pragma clang diagnostic pop
                             ^

/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/prim/err/check_matching_dims.hpp:33:8: warning: unused variable 'error' [-Wunused-variable]
  bool error = false;
       ^

/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/prim/err/hmm_check.hpp:33:7: warning: unused variable 'n_transitions' [-Wunused-variable]
  int n_transitions = log_omegas.cols() - 1;
      ^

/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/prim/fun/gp_matern52_cov.hpp:275:10: warning: unused variable 'neg_root_5' [-Wunused-variable]
  double neg_root_5 = -root_5;
         ^
 /Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/prim/fun.hpp:183:
/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/prim/fun/log_mix.hpp:86:13: warning: unused variable 'N' [-Wunused-variable]
  const int N = stan::math::size(theta);
            ^

/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/prim/prob/double_exponential_cdf.hpp:82:10: warning: unused variable 'N' [-Wunused-variable]
  size_t N = max_size(y, mu, sigma);
         ^

/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/prim/prob/gaussian_dlm_obs_rng.hpp:98:7: warning: unused variable 'n' [-Wunused-variable]
  int n = G.rows();  // number of states
      ^

/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/prim/prob/hmm_marginal.hpp:26:13: warning: unused variable 'n_states' [-Wunused-variable]
  const int n_states = omegas.rows();
            ^
/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/prim/prob.hpp:315:
/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/prim/prob/std_normal_rng.hpp:23:22: warning: unused variable 'function' [-Wunused-variable]
  static const char* function = "std_normal_rng";
                     ^
/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/prim/prob.hpp:335:
/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/prim/prob/von_mises_cdf.hpp:72:10: warning: unused variable 'ck' [-Wunused-variable]
  double ck = 50;
         ^

/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/src/stan/io/dump.hpp:778:15: error: no member named 'validate_dims' in namespace 'stan::io'
    stan::io::validate_dims(*this, stage, name, base_type, dims_declared);
    ~~~~~~~~~~^

/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/src/stan/model/indexing/rvalue.hpp:610:13: warning: unused variable 'cols' [-Wunused-variable]
  const int cols = rvalue_index_size(idxs.tail_.head_, x_ref.cols());
            ^

/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/src/stan/model/indexing/rvalue_varmat.hpp:241:22: warning: unused variable 'x_cols' [-Wunused-variable]
  const Eigen::Index x_cols = x.cols();
                     ^

/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/rev/core/operator_subtraction.hpp:88:43: warning: lambda capture 'b' is not used [-Wunused-lambda-capture]
                            [avi = a.vi_, b](const auto& vi) mutable {
                                          ^
/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/rev/core/operator_minus_equal.hpp:24:16: note: in instantiation of function template specialization 'stan::math::operator-<double, nullptr>' requested here
  vi_ = (*this - b).vi_;
               ^
/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/prim/core/complex_base.hpp:136:9: note: in instantiation of member function 'stan::math::var_value<double, void>::operator-=' requested here
    re_ -= x;
        ^
/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/prim/core/operator_subtraction.hpp:24:5: note: in instantiation of function template specialization 'stan::math::complex_base<stan::math::var_value<double, void> >::operator-=<int>' requested here
  y -= rhs;
    ^
/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/prim/core/operator_subtraction.hpp:55:20: note: in instantiation of function template specialization 'stan::math::internal::complex_subtract<std::__1::complex<stan::math::var>, int>' requested here
  return internal::complex_subtract(x, y);
                   ^
/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/prim/fun/acosh.hpp:105:31: note: in instantiation of function template specialization 'stan::math::operator-<stan::math::var_value<double, void>, int>' requested here
  auto y = log(z + sqrt(z * z - 1));
                              ^
/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/rev/fun/acosh.hpp:94:32: note: in instantiation of function template specialization 'stan::math::internal::complex_acosh<stan::math::var_value<double, void> >' requested here
  return stan::math::internal::complex_acosh(z);
                               ^

/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/rev/core/operator_addition.hpp:84:43: warning: lambda capture 'b' is not used [-Wunused-lambda-capture]
                            [avi = a.vi_, b](const auto& vi) mutable {
                                          ^
/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/rev/fun/grad_inc_beta.hpp:45:43: note: in instantiation of function template specialization 'stan::math::operator+<int, nullptr>' requested here
    grad_2F1(dF1, dF2, a + b, var(1.0), a + 1, z);
                                          ^

/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/rev/core/operator_addition.hpp:84:43: warning: lambda capture 'b' is not used [-Wunused-lambda-capture]
                            [avi = a.vi_, b](const auto& vi) mutable {
                                          ^
/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/prim/fun/trigamma.hpp:63:31: note: in instantiation of function template specialization 'stan::math::operator+<double, nullptr>' requested here
    value = -trigamma_impl(-x + 1.0) + square(pi() / sin(-pi() * x));
                              ^
/Library/Frameworks/R.framework/Versions/4.1/Resources/library/StanHeaders/include/stan/math/rev/fun/trigamma.hpp:23:44: note: in instantiation of function template specialization 'stan::math::trigamma_impl<stan::math::var_value<double, void> >' requested here
inline var trigamma(const var& u) { return trigamma_impl(u); }
brms stan bayesian r
1 Answers

I added this as an answer because it was just too much for the comment. I want to help you figure this out.

1) Versions

First, please compare the versions of brms and the dependencies. I know you wrote that you checked this, but these errors are odd! It can't hurt to compare them. It's probably important to note that I have a Mac (OS 12.2.1).

I used this code to gather the information; here are my versions.

# brms and it's dependencies
dPkg <- c("brms", "rstan", "ggplot2", "loo", "posterior", "Matrix", "mgcv", "rstantools",
          "bayesplot", "shinystan", "bridgesampling", "glue", "future", "matrixStats", 
          "nleqslv", "nlme", "coda", "abind", "stats", "utils", "parallel", "grDevices", 
          "backports")

cbind(dPkg, purrr::map(dPkg, ~as.character(packageVersion(.x))))
#       dPkg                      
#  [1,] "brms"           "2.16.3" 
#  [2,] "rstan"          "2.21.3" 
#  [3,] "ggplot2"        "3.3.5"  
#  [4,] "loo"            "2.4.1"  
#  [5,] "posterior"      "1.2.0"  
#  [6,] "Matrix"         "1.4.0"  
#  [7,] "mgcv"           "1.8.38" 
#  [8,] "rstantools"     "2.1.1"  
#  [9,] "bayesplot"      "1.8.1"  
# [10,] "shinystan"      "2.5.0"  
# [11,] "bridgesampling" "1.1.2"  
# [12,] "glue"           "1.6.1"  
# [13,] "future"         "1.24.0" 
# [14,] "matrixStats"    "0.61.0" 
# [15,] "nleqslv"        "3.3.2"  
# [16,] "nlme"           "3.1.155"
# [17,] "coda"           "0.19.4" 
# [18,] "abind"          "1.4.5"  
# [19,] "stats"          "4.1.1"  
# [20,] "utils"          "4.1.1"  
# [21,] "parallel"       "4.1.1"  
# [22,] "grDevices"      "4.1.1"  
# [23,] "backports"      "1.4.1"

2) expose_functions and rstan

I have found that expose_functions is a regurgitation of the function rstan::expose_stan_functions(), so I went there to look for answers.

I found this website. If you haven't seen it, you might find it pretty insightful. It doesn't help with the immediate issue, though.

In the rstan package, they had a straightforward example of this function:

library(rstan)
model_code <-
  '
  functions {
    real standard_normal_rng() {
      return normal_rng(0,1);
   }
  }
'
expose_stan_functions(stanc(model_code = model_code))

It doesn't give output, either (just like the issue I saw with expose_functions). I received a warning that it expected a list of integer vectors--but it worked. Does it work for you? If it does, we have to go back to the models.

I'm not sure what the real purpose of this function is. The output is the name of the function. So the output of this last bit is:

# [1] "standard_normal_rng"

And the output of expose_functions in your question is written in the code when stans_funs was created. This is the output of the original expose functions:

# [1] "beta_binomial2_lpmf" "beta_binomial2_rng"

When you run the library rstan, it tells you that you are recommended to add a few things to your script. I followed the recommendation.

options(mc.cores = parallel::detectCores()) # added as rec by lib rstan
rstan_options(auto_write = T)               # added as rec by lib rstan

3) Back to the models

If that call for the function from rstan works, then I would like to know what you got for the output of fit1 and fit2. I didn't think the entire output of summary() needs to be added, so I highlighted a few things from the model summaries:

Fit 1

  • family: binomial; link: mu = logit
  • draws: 4 chains w/ iter = 2K, warmup = 2K, thin = 1, post-warmup = 4K

Fit 2

  • family: beta_binomial2; links: mu = logit & phi = identity
  • draws are the same as fit1

As far as the loo function, that is derived from the package loo It may be better to wait to dive into that until the other parts are checked, eh?

user_53290730
Related