Skip to content

cannot increase the number of iterations in lme call over 146

5 messages · Antonio P. Ramos, Joshua Wiley

Original
Joshua Wiley
# 
Hi Antonio,

Look at the error message: "function evaluation limit reached without
convergence" you increased the number of iterations, but the objective
function reached its limit for max evaluations prior to convergance.
The 'brute force' approach would be to use the msMaxEval argument of
lmeControl to up that, but I would suggest carefully scrutinizing your
data and model prior before blindly asking the optimizer to run
longer.

Have you graphed your data?  What sort of variables are time and
maternal_educ?  If you send us data, we can show you some examples of
how you might graph and examine your data.  Also, do you really want a
random interaction?

Finally, in model formulae, * behaves specially, so x * z expands to:
x + z + x:z, thus a simplified writing of your model (not really
important and if the other way is clearer to you, by all means use it,
but you can save a few keystrokes; also note I show the msMaxEval
argument):

model.c2 <- lme(log(child_mortality) ~  time * log(maternal_educ),
  control = lmeControl(msMaxIter = 200, msMaxEval = 500 msVerbose = TRUE),
  merged1, random = ~ log(maternal_educ) * time | country.x,
  na.action = na.omit, method = "ML")

Cheers,

Josh

On Tue, Jan 3, 2012 at 5:52 PM, Antonio P. Ramos
<ramos.grad.student at gmail.com> wrote:

  
    

      
      
    

  


  


      

    

    

      
      

        
      

    

    

      
      

        


  

        

      Joshua Wiley
    

    

      
      #
    

  


  

      
On Tue, Jan 3, 2012 at 7:20 PM, Antonio P. Ramos

        
<ramos.grad.student at gmail.com> wrote:

        
[snip]
 at end of the day, it still don't undertand why it

            

      
      
      
    

        
It does increase the number of iterations---you reached a function
evaluation limit, not an iteration limit.  There is not a 1:1
relationship between function evaluations and iterations.  If you're
curious, you may enjoy: http://netlib.bell-labs.com/cm/cs/cstr/153.pdf
 or http://www.numerical-recipes.com/ the first is some documentation
for the PORT routines which are used by nlimnb(), the latter is goes
into more detail about optimization.  There are other (possibly much
better) books out there, but it is the only one I am personally
familiar with.

Cheers,

Josh