"Stan model 'negbin-glm' does not contain samples." error

[malai...@gmail.com]

When i run the attached code i am getting an error "Stan model 'negbin-glm' does not contain samples", please help clarifying why I am getting that error and how i can debug it.

Thanks.

# data processing

getwd()

setwd("C:/Users/n01066737/Desktop/Angela")

getwd()

crashdata<-read.csv(file.choose(), header = T)

crashdata

summary(crashdata)

names(crashdata)

str(crashdata)

### converting factor variables to as.factors

crashdata$time_indicator<-as.factor(crashdata$time_indicator)

crashdata$treatment<-as.factor(crashdata$treatment)

crashdata$median_type<-as.factor(crashdata$median_type)

crashdata$landuse<-as.factor(crashdata$landuse)

crashdata$speed<-as.factor(crashdata$speed)

### checking the structure after converting the factor variables

str(crashdata)

### loading the rstan package

library(rstan)

### creating dummy variables for factor variables

cat_time<-model.matrix(~crashdata$time_indicator, data=crashdata)

cat_treatment<-model.matrix(~crashdata$treatment, data=crashdata)

cat_median_type<-model.matrix(~crashdata$median_type, data=crashdata)

cat_landuse<-model.matrix(~crashdata$landuse, data=crashdata)

cat_speed<-model.matrix(~crashdata$speed, data=crashdata)

### specifying the poisson-gamma mixture model

negbin<-"

data{

int<lower=0> N;                           # rows of data

vector[N] cat_time;                       # predictor variable

vector[N] cat_treatment;                  # predictor variable

vector[N] cat_median_type;                # predictor variable

vector[N] cat_landuse;                    # predictor variable

vector[N] cat_speed;                      # predictor variable

vector[N] ln_maj_aadt;                    # predictor variable

vector[N] ln_min_aadt;                    # predictor variable

vector[N] no_lanes_maj;                   # predictor variable

vector[N] no_lanes_min;                   # predictor variable

vector[N] no_lanes_min;                   # predictor variable

vector[N] median_width;                   # predictor variable

int<lower=0> total_crashes[N];            # response variable

}

parameters {

real<lower=0> phi;      # negative binomial dispersion parameter

real beta0;             # intercept

real beta1;

real beta2;

real beta3;

real beta4;

real beta5;

real beta6;

real beta7;

real beta8;

real beta9;

real beta10;

}

model {

### priors

phi ~    cauchy(0.001,0.001);

beta0 ~  normal(0.0, 1.0E4);

beta1 ~  normal(0.0, 1.0E4);

beta2 ~  normal(0.0, 1.0E4);

beta3 ~  normal(0.0, 1.0E4);

beta4 ~  normal(0.0, 1.0E4);

beta5 ~  normal(0.0, 1.0E4);

beta6 ~  normal(0.0, 1.0E4);

beta7 ~  normal(0.0, 1.0E4);

beta8 ~  normal(0.0, 1.0E4);

beta9 ~  normal(0.0, 1.0E4);

beta10 ~ normal(0.0, 1.0E4);

# data model

total_crashes~ neg_binomial_2_log(beta0 + beta1*cat_time + beta2*cat_treatment + beta3*cat_median_type + beta4*cat_landuse + beta5*cat_speed + beta6*ln_maj_aadt + beta7*ln_min_aadt + beta8*no_lanes_maj + beta9*no_lanes_min +beta10*median_width, phi);

}

"

write(negbin, file="negbin-glm.stan")

sm<-stan_model("negbin-glm.stan")

set.seed(1000)

N<- 849

phi<-0.7349047  

beta0<--3.150702

beta1<-0.0479721

beta2<-0.3503469

beta3<-0.0028539

beta4<--0.1059735

beta5<-0.6936886 

beta6<-0.2642703

beta7<-0.121047

beta8<-0.0255357

beta9<-0.1235194

beta10<--0.0093531

expected_crashes<-sampling(sm, data=list(N=N, crashes=crashdata$total_crashes, cat_time=crashdata$time_indicator -mean((crashdata$time_indicator)), cat_treatment=crashdata$treatment-mean((crashdata$treatment)), cat_median_type=crashdata$median-type-mean((crashdata$median-type)), cat_landuse=crashdata$landuse-mean((crashdata$landuse)), cat_speed=crashdata$speed-mean((crashdata$speed)), ln_maj_aadt=crashdata$ln_aadt_maj-mean((crashdata$ln_aadt_maj)), ln_min_aadt=crashdata$ln_min_aadt-mean((crashdata$ln_min_aadt)), no_lanes_maj=crashdata$no_lanes_maj-mean((crashdata$no_lanes_maj)), no_lanes_min=crashdata$no_lanes_min-mean((crashdata$no_lanes_min)), median_width=crashdata$median_width-mean((crashdata$median_width))), pars=c("beta0", "beta1", "beta2", "beta3", "beta4", "beta5", "beta6", "beta7", "beta8", "beta9", "beta10", "phi"), iter=50000, warmup=20000, chains=1)

expected_crashes

# plot the output

rstan::traceplot(expected_crashes, inc_warmup=FALSE)

library(rstanarm)

library(shinystan)

launch_shinystan(expected_crashes)

[Bob Carpenter]

I'd suggest starting with a simpler example and working up.
Or you can try to use print statements in the existing program
to see where things go wrong if you're not getting a specific
line number out. Test with simulated data on the simpler thing
to make sure it works.

Usually sampling fails due to specific exceptions being
thrown in numerical methods that expect inputs in certain ranges
or will overflow and underflow.

Putting variables on very different scales, as in this:

phi ~ cauchy(0.001,0.001);

can be problematic. Ideally for speed and stability,
parameters posteriors will be roughly normal(0, 1)---otherwise,
we have to rescale during adaptation.

- Bob

> --
> You received this message because you are subscribed to the Google Groups "Stan users mailing list" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to stan-users+...@googlegroups.com.
> To post to this group, send email to stan-...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.
> <FULL-BAYES-DATA.csv><pcs_project_1.R>

[Lu Zhang]

I agree that the prior "phi ~ Cauchy(0.001,0.001)" could be problematic. It could be too informative.  See the following picture:

And I noticed that the name of your variable in the CSV file contain "." while all the variable names you use in the code have "_". It may cause problems 

Finally, I would recommend you use vector beta instead of setting coefficient one by one. You could use 

data{

    int<lower = 0> N;                               # rows of data

    int<lower = 0> P;                               # colums of design matrix

    matrix[N, P] X;                                  # design matrix

    int<lower=0> total_crashes[N];         # response variable

}

parameters {

real<lower=0> phi;      # negative binomial dispersion parameter

vector[P] Beta;

}

model {

### priors

phi ~    cauchy(0, 0.1);

for (i in 1:P) {

  Beta[i] ~ normal(0.0, 1.0E4);

}

# data model

total_crashes~ neg_binomial_2_log(X*Beta, phi);

}

Best,

Lu Zhang

[Angel Edes]

Thank you so much Bob, and Lu.

I will work on you're recommendations.

--
You received this message because you are subscribed to a topic in the Google Groups "Stan users mailing list" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/stan-users/J-o6ea9Ed4I/unsubscribe.
To unsubscribe from this group and all its topics, send an email to stan-users+unsubscribe@googlegroups.com.

[malai...@gmail.com]

Hi All,

I dropped all the variables in my code and remained with one variable, and when i run it again i received an error

"Error : variable does not exist; processing stage=data initialization; variable name=minaadt; base type=vector_d

failed to create the sampler; sampling not done"

What does this indicate?

Please help

[Bob Carpenter]

A variable named minaadt isn't getting a value in the data input
for the program.

- Bob

> --

> You received this message because you are subscribed to the Google Groups "Stan users mailing list" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to stan-users+...@googlegroups.com.

[malai...@gmail.com]

Thank you Bob,

I was able to figure out the problem.