Control parameters to reproduce ergm 3.11 (last in the 3.* series) estimation settings

[krivit]

This is a frequently requested feature, so I am opening a ticket here for how to provide this functionality.

My current idea is to implement a control.ergm3() function that simply passes its arguments to control.ergm() but whose defaults set up the estimation along the lines of 3.11. I've created a new branch, control3, that implements, documents, and tests such a function. Thanks to some R function definition alchemy, it should automatically track the unaffected parameters and implementation of control.ergm().

As of this writing, the two affected settings are:

• MCMLE.termination = "Hummel": Use Hummel (2012) termination criterion instead of confidence.
• MCMLE.effectiveSize = NULL: Disable adaptive MCMC.

@sgoodreau , @smjenness , @mbojan , @CarterButts , since you and your teams are some of the most frequent encounterers (an actual word) of situations in which problems that converged on 3.* no longer do no 4.*, can you please,

• Test if this works for you?
• Suggest any other settings that need to be adjusted?

[sgoodreau]

Thank you! I will reach out to Aditya, since it was his example that I started down this road for. I'll point him to this and see if he can do some testing.

@smjenness - when I was helping you to diagnose your edges+offset(isolates) model, there was a zone of network stats that I could get to converge in 3.x but not in 4.x. We subsequently discovered that the model itself is a bit of a challenge, but even then this should provide a useful test case. I will work on this one.