Perry de Valpine

Hi @danielturek @paciorek . First thanks for all of this. Here are quick thoughts to be followed by code review later. 1. Can we bounce around names? For the options...

@danielturek @paciorek Some code notes: 1. In the roxygen documentation strings for `getNodeNames`: The text implies posterior predictive nodes must be stochastic, but I think people use predictive nodes for...

Oops I wasn't done with that and my fingers posted it by accident. Let me edit it!

More comments: 1. MCMC_build 95-96: Would it work to save the result of `grepl('^posterior_predictive', samplerNames)` so we don't need to repeat it? 2. I might not be following completely but...

Here is a model case to discuss: A --> B --> C and also A --> C. C is posterior predictive. Samplers are A', B', C' with targets A, B,...

Two comments on efficiency and error-trapping: 1. It looks like during model initialization, called from `modelDefClass$newModel`, data will always start empty, so the call to `setPostPredNodeIDs` from `init_isDataEnv` from `initialize`...

@danielturek My initial scenario was not entirely clear, and the intention was what @paciorek has indicated: Only C is posterior predictive. If we agree this scenario is a problem, an...

@danielturek This idea makes sense to me. I like the simplification. We could perhaps add a control parameter to turn off downstream updating and only update the individual target node....

I'd welcome ideas on this. I suspect that a variable whose name is *any* C++ keyword will lead to failures. We could grab a [list of C++ keywords](http://en.cppreference.com/w/cpp/keyword) and warn...

Can you try second call with resetFunctions = TRUE. I think it’s part of larger need for more flexible compilation management. > On May 18, 2016, at 9:02 AM, Christopher...