1

I am having a problem when trying to summarise my aditive predictor:

mu[j] <- b0 +  weights1[1] * A[j] + weights1[2] * A[j+1] + weights1[3] * A[j+2] + weights1[4] * A[j+3] +
     weights1[5] * A[j+4] + weights1[6] * A[j+5] + weights1[7] * A[j+6] + weights1[8] * A[j+7] + 
     weights1[9] * A[j+8] + weights1[10] * A[j+9] + weights1[11] * A[j+10] + weights1[12] * A[j+11] + 

     weights2[1] * B[j] + weights2[2] * B[j+1] + weights2[3] * B[j+2] + weights2[4] * B[j+3] +
     weights2[5] * B[j+4] + weights2[6] * B[j+5] + weights2[7] * B[j+6] + weights2[8] * B[j+7] + 
     weights2[9] * B[j+8] + weights2[10] * B[j+9] + weights2[11] * B[j+10] + weights2[12] * B[j+11]

by using inprod(). This is what I thought should be the equivalent:

     mu[j] <- b0 + inprod(weights1[],A[j:(j+11)]) + inprod(weights2[],B[j:(j+11)])

While the model compiles and seems to work, it stays updating forever. Its been running for hours and it does not end while the first approach ends in few minutes.

These are the priors, just in case:

weights1[1] ~ dnorm(0,1.0E-6)
weights2[1] ~ dnorm(0,1.0E-6)
for(t in 2:12) {
weights1[t]~dnorm(weights1[t-1],tauweight1)}
for(t in 2:12) {
weights2[t]~dnorm(weights2[t-1],tauweight2)}
b0 ~ dnorm(0,.001)
tau ~ dgamma(0.001, 0.001)
sigma <- 1/sqrt(tau)
tauweight1~dgamma(1.0E-3,1.0E-3)
tauweight2~dgamma(1.0E-3,1.0E-3)

I am calling OpenBUGS from R using R2OpenBUGS just in case.

Thanks very much for your time!

IosuP
  • 11
  • 2
  • I have realised that the code does work. The problem is that it takes an extremely long tim when using **inprod()** (150s for 10 iterations). I guess its related to the fact that my weights are correlated through a first order random walk. Any suggestions on how to speed this up? – IosuP Jan 08 '19 at 13:17

0 Answers0