Daniel Turek

@paciorek Thanks for pointing that out. I'll address some of those in a message here later.

@perrydv @paciorek Let me describe (briefly) what this undertaking is, since I note it is not fully in-line with some of the thinking in NCT issues 285 and 269. Those...

[Daniel is in disbelief that testing has passed]

@paciorek I think this warrants some discussion, about the ordering MCMC samplers (w.r.t. PP nodes). No, at present the MCMC does nothing to put the sampling of PP nodes "last"....

@paciorek Very good discussion. I agree, your point about sampling parameters **unconditional** on the PP nodes does raise a different situation, outside of the standard Gibbs sampling I was thinking...

@paciorek @perrydv I'm reasonably happy with this PR. A quick description of the changes here: - The following new options are added to both `getNodeNames` and `getDependencies`: `includePosteriorPred`, `posteriorPredOnly`, `includePosteriorPredBranch`,...

@perrydv Thanks for such a thorough review, and the great suggestions. Trying to organize ideas here, I'm going to introduce letters to address different points. (a) I'm fine with any...

@perrydv @paciorek I've thought about this a while. I cannot convince myself that it's **not** a potential problem, although I tried a number of test cases (manually tweaking whether samplers...

One more thought / question about this: I took some care in the (current) implementation to ensure that when PP nodes are **excluded** from the sampler calculations, the **logProbs of...

@paciorek @perrydv What do you guys think about the following idea? I think it's a nice simplification. Eliminate the notion of "posterior predictive **branch points**", and also eliminate the `posterior_predictive_branch_sampler`....