buildMCEM gives spurious warning about unsampled nodes

[perrydv]

buildMCEM creates an MCMC for latent states but not parameters.

Currently a user sees a warning like:
[Warning] No samplers assigned for 5 nodes, use conf$getUnsampledNodes() for node names.

I believe this is emitted from configureMCMC when it is being called to configure only for the latent nodes.
It would be nice to silence this warning. It may be as simple as print=FALSE.

[perrydv]

Expanding this issue to other notes on buildMCEM:

• There is an error message involving forceNoConstraints, which is out of date. The error message should be revised.
• There is no feature for trying to pick default initial parameter values. We could do that similarly to buildLaplace.
• We should check for package mcmcse during setup code. Currently it errors out if it runs and doesn't find that.
• The initialization at the start of findMLE seems to be simulating parameters from their priors even though those priors won't be used and initial values are provided as an argument to findMLE.

[paciorek]

Ok, I silenced the warning in a local branch on my machine.

Regarding your other suggestions:

• Regarding forceNoConstraints, do you mean the message should indicate that that is available via control rather than a top-level argument to buildMCEM?
• Regarding initial parameter values, I'm not sure what you are referring to in terms of how Laplace does it. Laplace's findMLE has a pStart that looks the same to me as findMLE for MCEM.
• I put in a check for mcmcse in buildMCEM setup (also in local branch).
• Regarding inits for findMLE, I see if(any(abs(pStart) == Inf)) pStart <- values(model, paramNodes) in line 1221 of MCEM_build.R. I don't see any simulation anywhere in MCEM_build.R. Also, I can't parse your comment "even thought those priors won't be used an initial values are provided as an argument to findMLE".

[perrydv]

Thanks @paciorek . This was a hasty issue with some typos (some I edited) as I was noticing some things. I will make a PR.

[paciorek]

Ok, but note I already did a bit of work on this. See branch cleanup_mcem.

[perrydv]

Thanks @paciorek . See #1601 . This turned out to be a shallow rabbit hole with many things being noticed.

For the record here:

• Regarding initial parameter values, there was a bug of not using them correctly. Also what I meant was that Laplace initializes to a vector of zeros in the transformed space if nothing is provided. We don't have that for MCEM (and I did not add it at this time but we could).
• Yes, you are correct about forceNoConstraints. I modified a warning message to clarify that when we catch deprecated arguments, many of them were moved to be control list elements.
• I think my note about inits reflected other problems and was a hasty mis-interpretation of something I was seeing.