I have a parametrised linear Gaussian Bayesian network and I am trying to make predictions on the model using rjags. I can do this for one observation but do not know how to pass multiple observations. Here is an example
library(rjags)
library(coda)
Initial model
mod <- textConnection("model {
mpg.hat <- (34.96055404 - 3.35082533* wt - 0.01772474* disp)
wt ~ dnorm(3.21725, 1/0.9784574^2)
disp ~ dnorm(230.7219, 1/123.9387^2)
mpg ~ dnorm(mpg.hat, 1/2.916555^2)
}")
# Evaluate and get prediction when wt=1 and disp is hidden
m <- jags.model(mod, n.chains = 1, n.adapt = 1000, data=list(wt=1, disp=NA))
update(m, 10000)
cs <- coda.samples(m, c("mpg", "wt", "disp"), 1e5)
summary(cs)
This works as expected, however, I have multiple rows of data that I want to generate predictions for. If I try to extend the data=list( argument to include more rows it throws an error. So after rerunning the model text, and the following command I get the error
m <- jags.model(mod, n.chains = 1, n.adapt = 1000, data=list(wt=1:2, disp=1:2))
Error in jags.model(mod, n.chains = 1, n.adapt = 1000, data = list(wt = 1:2, :
Error in node dnorm(230.722,(a1/(a123.939^2)))
Length mismatch in Node::setValue
How do I extend this to more observations?
You need to iterate through the rows:
mod <- textConnection("model {
for (n in 1:N) {
mpg.hat[n] <- (34.96055404 - 3.35082533* wt[n] - 0.01772474* disp[n])
mpg[n] ~ dnorm(mpg.hat[n], 1/2.916555^2)
wt[n] ~ dnorm(3.21725, 1/0.9784574^2)
disp[n] ~ dnorm(230.7219, 1/123.9387^2)
}
}")
Note you will need to add N to your data list as well:
data = list(N = 1, wt = 1, disp = NA)
data = list(N = 2, wt = 1:2, disp = 1:2)