Removing rngR

[LeahPrice]

RNGscope is established automatically with Rcpp attributes, so the unnecessary rng files have been removed.

Also fixing up the move selector function so that it is on the particle level rather than population level.

[eddelbuettel]

Nice. Looks good to me.

Will wait on Adam to chime in too before merging.

[adamjohansen]

It's been a long day, so apologies if I'm missing something, but do we even need this helper function? (By which I mean can we not just call rmultinom directly when it's used which I think is just 393/403 for resampling). It's not a big deal either way but less code is so often more...

[eddelbuettel]

I was wondering this too this morning (but have been out all day -- still vacationing). As I recall we now have a multinomial routine elsewhere, ie in Rcpp's rmulitnom wrapping R's own.

But we can keep these cleanups for another time.

[adamjohansen]

Quite so -- cleanups can be done later -- I was just wondering if this was really needed and this seemed a good time to ask.

[LeahPrice]

Yeah I agree it would be better without that helper function. The Multinomial function uses rmultinom but it has a few extra lines to deal with the fact that uRSCount is of type arma::Col and rmultinom wants int *. I used uRSCount.memptr() to get type unsigned int * but I was having trouble casting to int *. Do you think it's an okay solution to just change uRSCount to type arma::Col? That's what I've tried and I'll do a pull request now for it. Feel free to reject it obviously if it's a bad solution.

[adamjohansen]

You can't (really) cast unsigned* to int* because there's no guarantee that int and unsigned int have a closely related interpretation. You could use reinterpret_cast to tell the compiler to pretend that an unsigned* is an int* which would produce code that compiles... but it would have undefined behaviour.

The only reason for using an unsigned type was that it was a better logical description of a number of replicates. If there are pragmatic advantages for using a signed type, as there seem to be, then indeed that seems sensible.

(Don't feel that you have to respond to every question with new code, by the way, I was just curious as to whether it was really necessary -- and as Dirk said, there's no reason that small things like this can't be tidied up later if it's desirable to.)

[LeahPrice]

Okay, in that case I should probably change RSIndices to type int in a future version too (I'm using uRSCount += arma::conv_to<arma::Col<int> >::from(RSIndices))

Thanks, I'll keep that in mind for future minor changes like this.

[adamjohansen]

This looks good to me, I've got little to say other than sorry for being a bit slow to get back to you.