Is there any way to use previous run output for initialization?

Help
VelocideX
2013-09-12
2013-09-12
  • VelocideX

    VelocideX - 2013-09-12

    Suppose I have samples from a previous JAGS run.

    How can I pass the last of these samples (or the mean, etc) to JAGS to initialise variables for a new run? I have many many variables so transcribing manually is too hard.

    Thanks!

     
  • Martyn Plummer

    Martyn Plummer - 2013-09-12

    Using the rjags package you can use the coef function to get the current state of any chain. Here is an example using the BLOCKER bugs example:

    > m <- jags.model("blocker.bug", data, inits, n.chains = 2, quiet=TRUE)
    > update(m, 3000)
      |**************************************************| 100%
    > coef(m, chain=2)
    $d
    [1] -0.2380885
    
    $delta
     [1] -0.117981074 -0.209540191 -0.097065004 -0.324211821 -0.002261029
     [6] -0.355814837 -0.571193962 -0.327327590 -0.385089076 -0.251702755
    [11] -0.235192262 -0.097494862  0.173934550 -0.006762717 -0.097974005
    [16] -0.315847308 -0.605264102 -0.115847918 -0.408291936  0.152284433
    [21] -0.503832347 -0.054519856
    
    $delta.new
    [1] 0
    
    $mu
     [1] -2.343209 -2.153912 -2.185737 -2.423854 -2.368968 -2.474559 -1.623658
     [8] -2.025789 -1.696715 -2.361026 -2.259855 -1.526192 -2.918440 -2.819129
    [15] -1.491852 -1.340488 -2.027257 -3.195235 -3.200288 -1.477025 -2.112073
    [22] -3.039956
    
    $tau
    [1] 21.42443
    

    With the command line interface, there is a command parameters to which does the opposite of parameters in, e.g.

    model in blocker.bug
    data in blocker-data.R
    load glm
    compile, nchains(2)
    parameters in blocker-init.R
    initialize
    update 3000
    parameters to blocker-newinit.R, chain(2)
    
     
  • VelocideX

    VelocideX - 2013-09-12

    Thanks Martyn - this is exactly what I needed! My model takes a long time to adapt from random starting values, so this should help me get second and later runs going much closer to the parameter modes.

    I've only been using JAGS for a week now and it's a fantastic program - you've saved me a lot of time.

     

Log in to post a comment.

Get latest updates about Open Source Projects, Conferences and News.

Sign up for the SourceForge newsletter:





No, thanks