CODEWITHSUNDEEP
rjagsrjags
ShilZoyer
ShilZoyer

Reputation: 21

JAGS/BUGS error - Possible directed cycle involving some or all of the following nodes

My dataset is about penalty kicks and contains 106 rows and the features are :

I would like to perform a multinomial logistic regression on this data to have a model for the kick direction depending on the two others. I am taking as example the Aligator BUGS example : http://www.openbugs.net/Examples/Aligators.html

My BUGS file is the following one :

    model
{

# PRIORS
   alpha[1] <- 0; # zero contrast for baseline food
   for (k in 2 : K) {
      alpha[k] ~ dnorm(0, 0.00001) # vague priors
   }
# Loop around Foot:
   for (k in 1 : K){
      beta[1, k] <- 0
   } # corner-point contrast with first foot
   for (i in 2 : I) {
      beta[i, 1] <- 0 ; # zero contrast for baseline foot
      for (k in 2 : K){
         beta[i, k] ~ dnorm(0, 0.00001) # vague priors
      }
   }
# Loop around Time:
   for (k in 1 : K){
      gamma[1, k] <- 0 # corner-point contrast with first Time
   }
   for (j in 2 : J) {
      gamma[j, 1] <- 0 ; # zero contrast for baseline Time
      for ( k in 2 : K){
         gamma[j, k] ~ dnorm(0, 0.00001) # vague priors
      }
   }

# LIKELIHOOD   
   for (i in 1 : I) { # loop around Foot
      for (j in 1 : J) { # loop around Time

        # Multinomial response
         X[i,j,1 : K] ~ dmulti( p[i, j, 1 : K] , n[i, j] )
         n[i, j] <- sum(X[i, j, ])
         for (k in 1 : K) { # loop around Kick_Direction
            p[i, j, k] <- phi[i, j, k] / sum(phi[i, j, ])
            log(phi[i ,j, k]) <- alpha[k] + beta[i, k] + gamma[j, k]
            }
        }
    }
}

I use rjags and have the following error occuring :

Error in jags.model("kick_dir.bug", data, inits) : RUNTIME ERROR:
Possible directed cycle involving some or all
of the following nodes:
X[1,1,1:3]
X[1,2,1:3]
X[2,1,1:3]
X[2,2,1:3]
n[1,1]
n[1,2]
n[2,1]
n[2,2]

What did I do wrong ?

Thanks in advance

Upvotes: 2

Views: 251

Answers (1)

garej
garej

Reputation: 565

I had similar problem with dmulti distribution because it seems natural to code in this way by design.

The problem is in these two lines:

X[i,j,1 : K] ~ dmulti( p[i, j, 1 : K] , n[i, j] )
n[i, j] <- sum(X[i, j, ])

JAGS knows that the same variables X appear in LHS and RHS of the expression, which is prohibited in this package.

As one of possible workarounds, you could feed n[i, j] as part of data.

There is no such issue in STAN, by the way, as those sums get calculated automatically.

Upvotes: 0

Related Questions