mnrfit convergece

Hello,

I have been having some difficulty getting mnrfit to converge on certain iterations of my program:

Warning: Maximum likelihood estimation did not converge. Iteration limit exceeded. You may need to merge categories to increase observed counts. > In mnrfit>nominalFit at 496 In mnrfit at 193 In ComputeTrainingMetrics at 23 In DeepaktestmainJan27 at 40 In run at 57

At each iteration, observations are added to the mnrfit predictor matrix (X). Curiously, the convergence failure only happens after several iterations in which mnrfit converges successfully, meaning that the non-converging runs actually have more observations than the converging ones. This seems to go against what the warning message suggests to do.

What could explain this behavior?

Thanks!

Matt