Understanding Non-standard evaluation in 0.5.1

[bsaul]

The updates in v0.5.1 break my code, and I trying to understand why. I'm simulating a space-time structure, so I have two indexes s and t. simcausal understands a single dimension, so I map s and t to a single dimension via the function tFUN.

For each node in the DAG, I have a dataset of parameters that I lookup for s, t.

Here's a short example:

for(t in t_start:TT ){ # Per month
    for(s in s_start:SS ){ # Per location

      tpos  <- tFUN(s, t, SS)
      ## Initial month values ##

      if(s == 1 & t == 0){
       D <- D + node('L1',
               distr = 'rnorm',
               mean  = .(lookup_parameter(pp, node = 'L1', s = s, t = t, variable = '(Intercept)') ),
               sd    = .(lookup_parameter(pp, node = 'L1', s = s, t = t, variable = 'mse') ),
               t     = tpos)
 }
}

lookup_parameter basically uses [ to pick off the correct element in the dataset.

With the new updates, the values for s and t getting passed to lookup_parameter are not what I expected, and I get a One of the distribution parameters evaluated to non-standard vector error. t is the value I'd expect from simcausal rather then the value from the loop. s is a mystery altogether. The other arguments (node and variable pass to lookup_parameter correctly).

I think for my purposes (I'm trying to wrap up this study), I should just roll back to v0.5.0. I figured I would bring this up, in case there is something obvious I'm missing.

Thanks!

[osofr]

Hi,

Indeed, I highly recommend rolling back to 0.5.0 if you need to finish this quickly. This is quite a complex issue, and I have been struggling with how to do NSE in the right way for some time now. In version 0.5.0, anything inside .() was evaluated immediately upon calling the node() function. However, this had some undesirable consequences and starting from 0.5.1, the .() calls are evaluated only when the data is actually being simulated (i.e., when calling set.DAG() or sim() functions). Clearly this leads to incorrect interpretation of s and t, which should not be happening. I am looking into it right now.

Is it possible to provide a fully working example, with tFUN and lookup_parameter function? It will help me a great deal. Thanks!

Oleg

[osofr]

Would you mind trying to run your code on the most recent github commit? I've rolled back the non-standard evaluation with .() to be essentially the same as before. Please let me know if your old code works and if I can close the issue. Thanks!

[bsaul]

Sure thing. Will do tomorrow afternoon.

Sent from my iPhone

On Jun 1, 2016, at 20:33, Oleg Sofrygin notifications@github.com wrote:

Would you mind trying to run your code on the most recent github commit? I've rolled back the non-standard evaluation with .() to be essentially the same as before. Please let me know if your old code works and if I can close the issue. Thanks!

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[bsaul]

I installed v0.5.1.99 from github and everything appears to run fine. Thanks!

Do you still want me to provide a working example of the issue with version 0.5.1? I can, but won't be able to get to that till next week.

[osofr]

No need. Thanks for running it!