Technical documentation for pseudo likelihood iterations as implemented in fitglme

Hi all,

I'm currently trying to better understand the behavior and optimization algorithm implemented in fitglme, but I am finding the documentation lacking (either that, or I am unaware of where to look for more information).

Currently, when fitglme is running with "MPL" as the optimization method, the optimizer iterates over a series of steps until the tolerance criterion are met. This is all clear to me. However, at the end of a series of steps, a new MPL iteration occurs. I assume, between iterations the model parameters are modified in some way. Where can I find further information on 1.) how the initial model fitting parameters are set before the first iteration and 2.) how the model fitting parameters are modified between pseudo likelihood iterations?

Even with the optimizer running at a maximum verbosity of 2 it doesn't seem to print the fitting parameters to the console on each iteration.

Thanks in advance for your help,

Joseph

Answer 1

[Soumya]

Hi,

You can refer to the following documentation for some insightful information on fitglme

https://www.mathworks.com/help/stats/fitglme.html

Here, you can find examples which might help you.

As picked up from the documentation, the following line of code can be used to set the model specifications, where you can specify various parameters such as “Distribution”, “Link”, “FitMethod”, “StartMethod”, and many more:

glme = fitglme(mfr,'defects ~ 1 + newprocess + time_dev + temp_dev + supplier + 1|factory)','Distribution','Poisson','Link','log','FitMethod','Laplace','DummyVarCoding','effects')

To influence the initial parameter values, you can set the “StartMethod” to either "default" or "random". Here, the “default” means an internally defined default value and “random” is some randomly generated initial value.

Once the model is fitted you can use commands like “glme.Coefficients”, “glme.ModelCriterion” to get additional details on model fit coefficients, statistics and other properties.

If convergence issues arise, you can set the optimizer settings using “OptimizerOptions” where you can control the “MaxIter” values for the main iterations.

The documentation provides in-depth details, so I would recommend exploring the examples and other information to better understand the functionalities.