Reproducibility
Now that you've completed (and passed!) some Markov chain diagnostics, you're ready to finalize your RJAGS
simulation. To this end, reproducibility is crucial. To obtain reproducible simulation output, you must set the seed of the RJAGS
random number generator. This works differently than in base R
. Instead of using set.seed()
, you will specify a starting seed using inits = list(.RNG.name = "base::Wichmann-Hill", .RNG.seed = ___)
when you compile your model.
This is a part of the course
“Bayesian Modeling with RJAGS”
Exercise instructions
Run the provided code a few times. Notice that the
summary()
statistics change each time.For reproducible results, supply the random number generator
inits
tojags.model()
. Specify a starting seed of 1989.Run the new code a few times. Notice that the
summary()
statistics do NOT change!
Hands-on interactive exercise
Have a go at this exercise by completing this sample code.
# COMPILE the model
sleep_jags <- jags.model(textConnection(sleep_model), data = list(Y = sleep_study$diff_3))
# SIMULATE the posterior
sleep_sim <- coda.samples(model = sleep_jags, variable.names = c("m", "s"), n.iter = 10000)
# Summarize the m and s chains of sleep_sim
summary(sleep_sim)