Errors related to nimble Function: "dynamic index out of bounds" and "missing value where TRUE/FALSE needed"

[Jaume Adria Badia Boher]

Hi there!

I'm stuck in building a CMR multistate model with a spatial component and I would be very grateful if somebody could provide some help. 

My first aim is to generate dispersal distances for all individuals in my study split by sex on the basis of a gamma distribution. Some individuals have observed dispersal distances, but it has not been observed for some others. Therefore, the "Dist" vector is a combination of observed values and NAs.

for(i in 1:nind){

    Dist[i] ~ dgamma(sh[sexvec[i]], ra[sexvec[i]]) #Sexvec: vector indicating the sex of #each individual

  }

  

  for (g in 1:2){ #Priors

    sh[g] ~ dgamma(1,1)

    ra[g] ~ dgamma(1,1)

  }

The second step of the model would be to compare the dispersal distances of each individual with available recruitment points (which are known with certainty) and allocate every individual to the closest point to each dispersal distance. To do so, I entered an object to the model indicating the relative distances between each birth location and each potential recruitment point (Distmat), and a second object indicating the ID of the birth point of each individual (birthterr). Then, this part of the model must assign every individual to the closest recruitment point and return the ID of the recruitment point of each individual. 

Perry de Valpine kindly provided some feedback of this approach and suggested a function a few weeks ago: 

sortbyterr <- nimbleFunction(

  run = function(Distmat = double(2), birthterr = double(1), Dist = double(1), nterr = double()){

    returnType(double(1))

    n1 <- length(Dist)

    recterr <- numeric(length=n1, init=FALSE)

    

    for(i in 1:n1) {

      min_val <- Inf

      min_ind <- 1

      for(j in 1:nterr) { #nterr just indicates the total number of available territories

        this_val <- abs(Distmat[birthterr[i], j] - Dist[i])

        if(this_val < min_val) {

          min_val <- this_val

          min_ind <- j

        }

      }

      recterr[i] <-  min_ind

    }

    

    return(recterr)

  })

The function is incorporated into the model, and a second kernel (newkernel) is later calculated with the distances between birth and recruitment points.

recterr[1:nterr] <- sortbyterr(Distmat[1:51, 1:nterr], birthterr[1:nind], Dist[1:nind], nterr)

  

  for(i in 1:nind){

    newkernel[i] <- Distmat[birthterr[i],recterr[i]] 

  }

However, it seems there might be a problem in which the output of the function ("recterr") might not be created successfully. When compiling, I am warned that "dynamic index out of bounds: Distmat[birthterr_oBi_cB, recterr[i]]".

If that helps, Distmat is a 51*132 matrix and birthterr is length 461 with values between 1 and 51. 461 is the number of individuals in my study, 132 the number of points of potential recruitment, 51 is the number of points of birth.

Further errors are reported when trying to run the model.

Error in if(this_val < min_val){:

missing value where TRUE/FALSE needed

Note: cannot calculate logProb for node recterr[1:132]

Error in mapCopy. Sizes don't match: 132 != 461. 

The model will ultimately crash. I've been trying with other equivalent structures that incorporate the resulting "recterr" object and all provide errors I cannot identify as well. 

An alternative approach I've been trying to get this recterr object is to work with different built-in functions outside nimbleFunctions. Once the Dist object is calculated exactly identically to the first time, I1) calculate the difference between all distances in Distmat and the Dist object, 2) save the minimum value of each row, 3) use the built-in function "equals" to assign a  value of 1 to the row with the minimum value, and 0s elsewhere, and 4) use the built-in function "inprod" to obtain the territory number of the closest territory to the simulated Dist value.

for(i in 1:nind){

for(j in 1:nterr){

diffmat[i,j] <- abs(Distmat[birthterr[i],j] - Dist[i])

}

}

for(i in 1:nind){

minvec[i] <- min(diffmat[i,1:nterr])

}

for(i in 1:nind){

for(j in 1:nterr){

e[i,j] <- equals(diffmat[i,j], minvec[i])

}

}

for(i in 1:nind){

recterr[i] <- inprod(e[i,1:nterr],1:nterr)

}

Here my model complains about e[i,j] calculated by function equals, with several warnings indicating that:

Error in if (x1 == x2) 1 else 0 : 

  missing value where TRUE/FALSE needed

Note: cannot calculate logProb for node e[(...), (...)].

The model will crash as well at some point following this structure. Is it something wrong with my usage of the equals function?

My plan is to structure survival and recapture by point of recruitment in further steps, so I'll need to use this object as a dynamic index many times too.

Does anybody have any idea why either of the approaches fail? Any guess would be very appreciated. I tried to be as informative as possible but please let me know if you need any more information or remarks.

Thank you guys and have a nice day!

Jaume A.

[Perry de Valpine]

Hi Jaume,

You said the error "dynamic index out of bounds: Distmat[birthterr_oBi_cB, recterr[i]]" occurs when *compiling* the model. Can you clarify if it occurs during nimbleModel (which is not compiling, in our terminology) or in compileNimble (which is compiling)?

If the former, then the error is occuring during uncompiled (pure R) execution that is done to check the model after building it.  In that case, you can use typical debugging and error-trapping tools in R to track down the problem.

If you do "debug(sortbyterr)", you will get a browser every time the function is called, and you can inspect its inputs.  During uncompiled execution, this is like any other R function.

If instead of entering a browser every time, you want to catch an invalid value before it is returned, you could put something like this before the return statement:

if(any(recterr < 1) || any(recterr > 51)) {

   print("Got it!")

   browser()

}

You will then get into a browser where, with luck, you can inspect the inputs and determine what went wrong.  It looks like at some point either this_val or min_val is not well formed.

You will get warnings from nimbleFunction and/or nimbleModel about "browser" not being a nimbleFunction, but that is fine because you are not planning to compile the model with this piece of code.

The error involving "mapCopy" suggests that somewhere an index range or size provided is not what you want.  This could potentially happen if something in data, inits or constants is not the right size, or if there is a typo in model code.

I hope those ideas help in tracking it down!

-Perry

--
You received this message because you are subscribed to the Google Groups "nimble-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to nimble-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/nimble-users/a8d8605f-ade3-4298-a73c-41e117e1bcfen%40googlegroups.com.

[Jaume Adria Badia Boher]

Hi Perry,

and thanks for the feedback. I finally managed to move on following your suggestions on debugging.

Thanks again!

El dia dimecres, 21 de juliol de 2021 a les 18:31:27 UTC+2, pdevalpine va escriure:

[Perry de Valpine]

Great!

To view this discussion on the web visit https://groups.google.com/d/msgid/nimble-users/df25809a-2347-4dee-babd-88946fc902f3n%40googlegroups.com.