R Package: rube

Rube runs well on a Macintosh, but you must use jags, not WinBUGS. As of the beginning of July 2014, the creators of jags have not released a Mavericks version, so you must run the Snow Leopard version of R on your Mavericks machine.

Note for Macintosh Yosemite:
If you get an error message like:
jags failed: call: dyn.load(file, DLLpath = DLLpath, ...) error: unable to load shared object '/Library/Frameworks/R.framework/Versions/3.1/Resources/library/tcltk/libs/tcltk.so': dlopen(/Library/Frameworks/R.framework/Versions/3.1/Resources/library/tcltk/libs/tcltk.so, 10): Library not loaded: /opt/X11/lib/libX11.6.dylib Referenced from: /Library/Frameworks/R.framework/Versions/3.1/Resources/library/tcltk/libs/tcltk.so Reason: image not found
then you have run in to the fact the Yosemite changes where X11 and Quartz are installed relative to earlier OS10 versions. The quick fix is to add

, progress.bar="none"

Instructions for running rube on a Macintosh using a Windows emulator or CMU's Virtual Andrew are included in the main rube documentation file.

Rube runs fine from within R-Studio on either a PC or a Mac. The R-Studio limitation is that interactive graphics lik p3() cannot be "zoomed".

write(ali$model1, file="aliModel.txt")

Here are examples of how to use rube based on the "Aligators" example that comes with WinBUGS:

### Find any example as a ".odc" file in the WinBUGS example directory.
### Load the model, data, and initializations from that file:
library(rube)
ali <- getBugsExample("Aligators")
names(ali)  #  this example has only one model and one data set and one initialization.

### Show a summary of the model:
with(ali, summary(rube(model=model1, data=data1, inits=inits1)))
# No problems were found, and a nice summary of the model is produced:
## Rube Summary:
## For variables: i, j, k 
## 
## Assignments (logicals):
## alpha[], beta[,], gamma[,], mu[,,], b[,], g[,] 
## 
## Constants:
##   Size Min Max Mean SD NAs
## I    1            4      0
## J    1            2      0
## K    1            5      0
## 
## Data:
##   Distr  Size NAs Parameters (mean,sd) Initial Value(s) [Range] Flags
## X dpois 4x2x5   0            mu[i,j,k]  5.475 +/- 5.588 [0, 23]      
## 
## Stochastics:
##        Distr Size NAs  Parameters (mean,sd) Initial Value(s) [Range] Flags
## alpha  dnorm    5   1 0, 1e-05 (0, 316.228)           0 +/- 0 [0, 0]      
## beta   dnorm  4x5   8 0, 1e-05 (0, 316.228)           0 +/- 0 [0, 0]      
## gamma  dnorm  2x5   6 0, 1e-05 (0, 316.228)           0 +/- 0 [0, 0]      
## lambda dnorm  4x2   0 0, 1e-05 (0, 316.228)           0 +/- 0 [0, 0]      
## 
## No problems detected!


### Run the model through bugs by specifying parameters-to-save.
result = with(ali, rube(model1, data1, inits1, 
                        parameters.to.save=c("alpha","beta"), n.chains=1))
summary(result)
## Rube Summary:
## Run at 2010-08-10 08:04 and taking 2.98 secs 
## Based on 1000 kept iterations (n.thin=1) after burn-in of 1000 iterations
##             mean    sd MCMCerr   2.5%   25%   50%    75%    97.5%
## alpha[2]  -1.847 0.509  0.0690 -2.885 -2.14 -1.82 -1.496 -0.94746
## alpha[3]  -2.650 0.811  0.1017 -4.779 -3.01 -2.56 -2.135 -1.36475
## alpha[4]  -2.104 0.554  0.0341 -3.281 -2.47 -2.09 -1.708 -1.05980
## alpha[5]  -0.772 0.377  0.0327 -1.507 -1.01 -0.78 -0.528 -0.00548
## beta[2,2]  2.754 0.605  0.0776  1.688  2.33  2.68  3.139  3.98512
## beta[3,2]  3.012 0.618  0.0795  1.972  2.54  2.94  3.427  4.27110
## beta[4,2]  1.759 0.596  0.0732  0.679  1.34  1.72  2.126  3.02617
## 
## DIC = 191.816

p3(result) # produces graphical summaries


### bugs() wants a function for the initializations if you are 
### running multiple chains:
ali$inits
# $alpha
# [1] NA  0  0  0  0
# ...
simpleInit = function() {
  init = ali$inits
  init$alpha[2:5] = rnorm(4, 0, 1)
  return(init)
}
result3 = with(ali, rube(model1, data1, simpleInit, 
                         c("alpha","beta","gamma"), 
                         n.burn=1000, n.iter=9000))
p3(result3)

### Note that rube helps you document your models:
result3$startTime
result3$runTime
result3$model

### Demonstrate problem detection:
write(ali$model1, "aliModel.txt")
### Now, screw something(s) up in aliModel.txt, e.g.
### change one alpha[k] to alpha[k,m] and change a ")" to "("
### and misspell a function or density.
summary(rube("aliModel.txt", ali$data1, ali$inits1))
## You'll get something like this:
## Rube Summary:
## 
## No assignments.
## 
## No constants.
## 
## No data.
## 
## Stochastics:
## None
## 
## Problems:
## At line 12 in $model: Distribution must be ~dfun(...) or ~dfun(...)I(...). 
## beta[i, k] ~ dnorm(0, 0.00001( 
## At line 28 in $model: rpois is not on the list of valid distributions 
## X[i, j, k] ~ rpois(mu[i, j, k])

### Fixing errors may take multiple passes.  For my version of the above
### example, fixing the two errors then re-running rube() gives:
## Problems:
## Variable(s) with inconsistent dimensions: alpha