Dimensions of dynamically indexed variable

[Dylan Hubl]

Hi,

After updating to NIMBLE version 1.4.2, I've picked up a warning on code that I wasn't getting with an older version related to the dimensions of a variable that is dynamically indexed in my model. 

"[Warning] dimensions specified are larger than model specification for variable 'X'."

I am supplying variable 'X' as a matrix with 8 rows and 3 columns and passing it to the model as data. The model uses dynamic indexing to determine which row to use and I fix the column number in three different linear equations. Simplified example below to visualize:

for (j in 1:nsites[2]) { #season 2 with autoregression function
   
    f1[2,j] <- phi.s*X[z[1,j],1] + ....
    f2[2,j] <- phi.c*X[z[1,j],2] + ....
    f3[2,j] <- phi.b*X[z[1,j],3] + ....

In an attempt to determine the dimensions of the model specification of the variable, I have tried reducing the number of rows and columns in the matrix that I pass to the model. When I manually decrease the number of columns to two it throws an error "dimensions specified are smaller than model specification for variable 'X'." (as expected). However, I can manually reduce the number of rows down to two and still get the warning that the dimensions are larger than model specification. It's only when I manually supply a matrix with a single row and three columns that I get the error:

" Variable 'X' is dynamically-indexed but has at least one dimension of length one, probably because NIMBLE could not automatically determine its dimensionality. Please provide dimensions via the dimensions argument".

I'm wondering if this warning is popping up simply because it is dynamically indexed and the model isn't able to determine the dimensions during the model building phase. The model runs seemingly without issue with the warning.

Thanks,

Dylan

[Chris Paciorek]

Thanks for this report. I am able to reproduce this in nimble 1.4.2 and nimble 1.3.0 but not if I go back to nimble 1.1.1. 

We'll investigate. 

As you suggest, I don't think this will cause problems in using the model. 

-chris

--
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 visit https://groups.google.com/d/msgid/nimble-users/6862693c-bd25-441b-8294-a3e8be0a92adn%40googlegroups.com.