Compiling error in example on page 95 of Stan Reference Manual v. 2.2.0

[Edwin Leb]

Dear Community,

I am trying to run the example model for a multivariate (hierarchical) regression on page 95 of the Reference Manual for Stan version 2.2.0 but I am running into a compiling error.

This is the error message returned when calling stan():

TRANSLATING MODEL 'fullmodel' FROM Stan CODE TO C++ CODE NOW.

Error in stanc(file = file, model_code = model_code, model_name = model_name,  : 

  failed to parse Stan model 'fullmodel' with error message:

EXPECTATION FAILURE LOCATION: file=input; line=27, column=57

beta[j] ~ multi_normal((u[j] * gamma)’, Sigma_beta);

                                                        ^-- here

DIAGNOSTIC(S) FROM PARSER:

no matches for function name="multi_normal_log"

    arg 0 type=vector

    arg 1 type=row vector

    arg 2 type=matrix

available function signatures for multi_normal_log:

0.  multi_normal_log(vector, vector, matrix) : real

unknown distribution=multi_normal

Parser expecting: <eps>

I have actually slightly modified the code to fit a logistic rather than a linear model but I doubt this is the problem. Here is the modified code (note commented out lines):

fullmodel<- '

data {

int<lower=0> N; // num individuals

int<lower=1> K; // num ind predictors

int<lower=1> J; // num groups

int<lower=1> L; // num group predictors

int<lower=1,upper=J> jj[N]; // group for individual

matrix[N,K] x; // individual predictors

matrix[J,L] u; // group predictors

vector[N] y; // outcomes

}

parameters {

corr_matrix[K] Omega; // prior covariance

vector<lower=0>[K] tau; // prior scale

matrix[L,K] gamma; // group coeffs

vector[K] beta[J]; // indiv coeffs by group

#real<lower=0> sigma; // prediction error scale of the original linear model (commented out)

}

model {

matrix[K,K] Sigma_beta;

Sigma_beta <- diag_matrix(tau) * Omega * diag_matrix(tau);

tau ~ cauchy(0,2.5);

Omega ~ lkj_corr(2);

for (l in 1:L)

gamma[l] ~ normal(0,5);

for (j in 1:J)

beta[j] ~ multi_normal((u[j] * gamma)’, Sigma_beta);

for (n in 1:N)

#y[n] ~ normal(x[n] * beta[jj[n]], sigma); // original linear predictor commented out

y[n] ~ bernoulli(inv_logit(x[n] * beta[jj[n]]));

}

'

I am using R version 3.0.2 (2013-09-25) in Platform: x86_64-w64-mingw32/x64 (64-bit) and rstan_2.2.0.

Info on the compiler is:

> system("g++ --version")

G__~1.EXE (GCC) 4.6.3 20111208 (prerelease)

Is this really an error in the manual? Or what can I be doing wrong?

Thanks for any guidance.

Edwin

P.S. Explanations for beginners are highly appreciated.

[Soga Shota]

your error is in data block.

when you sample a random number from bernoulli distribution, the left hand side should be int. However, your code is

vector[N] y; // outcomes

which implicitly assume real values.

Therefore, you should modify your code like

data {

int<lower=0> N; // num individuals

int<lower=1> K; // num ind predictors

int<lower=1> J; // num groups

int<lower=1> L; // num group predictors

int<lower=1,upper=J> jj[N]; // group for individual

matrix[N,K] x; // individual predictors

matrix[J,L] u; // group predictors

# you have a typo here.

#vector[N] y; // outcomes

int y[N];

}

The error massage is kinda misreading. I spent a little bit of time to debug your code.

[Bob Carpenter]

> The error massage is kinda misreading. I spent a little bit of time to debug your code.b

Which error message? We might be able to make it clearer.

The one originally reported seemed clear enough in terms
of arguments to multi_normal. But it sounds like you were
responding to a different issue.

- Bob

[Edwin Leb]

Hi,

You are both right. I´ll recap to clarify.

There were two errors in my code. I reported the first one in my original mail was it was related indeed to the arguments to multi-normal. It seems I was using an unrecognized prime character (by R or stan? I don't know) to transpose (u[j] * gamma). I solved it by chance after unsuccessfully trying the change suggested by Soga ("the idea fell form sky"... hope the expression translates correctly to English).

Soga detected the second error in my code which would have popped as an error if I would have not changed the code as Soga suggested. 

In short the error messages thrown by stan seem correct. 

At the end I copy the error message I get when I run the model with the first problem corrected but without fixing the definition of y.

Many thanks for your input. Now I can go on exploring stan.

Edwin

Error for incorrectly defined data:

TRANSLATING MODEL 'Full_poly' FROM Stan CODE TO C++ CODE NOW.

Error in stanc(file = file, model_code = model_code, model_name = model_name,  : 

  failed to parse Stan model 'Full_poly' with error message:

EXPECTATION FAILURE LOCATION: file=input; line=30, column=53

y[n] ~ bernoulli(inv_logit(x[n] * beta[jj[n]]));

                                                    ^-- here

DIAGNOSTIC(S) FROM PARSER:

no matches for function name="bernoulli_log"

    arg 0 type=real

    arg 1 type=real

available function signatures for bernoulli_log:

0.  bernoulli_log(int, real) : real

1.  bernoulli_log(int, real[1]) : real

2.  bernoulli_log(int, vector) : real

3.  bernoulli_log(int, row vector) : real

4.  bernoulli_log(int[1], real) : real

5.  bernoulli_log(int[1], real[1]) : real

6.  bernoulli_log(int[1], vector) : real

7.  bernoulli_log(int[1], row vector) : real

unknown distribution=bernoulli

Parser expecting: <eps>

[Soga Shota]

It is my bad. An error massage was appropriate.

I fixed both errors at the same time (I should have mentioned both of them in my post) while I only saw the error from a wrong prime (and I thought an error in data section was not displayed... It was displayed if only a prime error was fixed).

Sorry for my misunderstanding.