RNG impact on MCMC [or lack thereof]

Following the talk at MCM 2017 about the strange impact of the random generator on the outcome of an MCMC generator, I tried in Montréal airport the following code on the banana target of Haario et al. (1999), copied from Soetaert and Laine and using the MCMC function of the FME package:

library(FME)
Banana <- function (x1, x2) {
 return(x2 - (x1^2+1)) }
pmultinorm <- function(vec, mean, Cov) {
 diff <- vec - mean
 ex <- -0.5*t(diff) %*% solve(Cov) %*% diff
 rdet <- sqrt(det(Cov))
 power <- -length(diff)*0.5
 return((2.*pi)^power / rdet * exp(ex)) }
BananaSS <- function (p) {
 P <- c(p[1], Banana(p[1], p[2]))
 Cov <- matrix(nr = 2, data = c(1, 0.9, 0.9, 1))
N=1e3
ejd=matrix(0,4,N)
RNGkind("Mars")
for (t in 1:N){
  MCMC <- modMCMC(f = BananaSS, p = c(0, 0.7), 
  jump = diag(nrow = 2, x = 5), niter = 1e3)
  ejd[1,t]=mean((MCMC$pars[-1,2]-MCMC$pars[1,2])^2)}

since this divergence from the initial condition seemed to reflect the experiment of the speaker at MCM 2017. Unsurprisingly, no difference came from using the different RNGs in R (which may fail to contain those incriminated by the study)…

MCM 2017 snapshots [#2]

On the second day of MCM 2017, Emmanuel Gobet (from Polytechnique) gave the morning plenary talk on regression Monte Carlo methods, where he presented several ways of estimating conditional means of rv’s in nested problems where conditioning involves other conditional expectations. While interested in such problems in connection with ABC, I could not see how the techniques developed therein could apply to said problems.

By some of random chance, I ended up attending a hard-core random generation session where the speakers were discussing discrepancies between GNU library generators [I could not understand the target of interest and using MCMC till convergence seemed prone to false positives!], and failed statistical tests of some 64-bit Mersenne Twisters, and low discrepancy on-line subsamples of Uniform samples. Most exciting of all, Josef Leydold gave a talk on ratio-of-uniforms, on which I spent some time a while ago  (till ending up reinventing the wheel!), with highly refined cuts of the original box.

My own 180 slides [for a 50mn talk] somewhat worried my chairman, Art Owen, who kindly enquired the day before at the likelihood I could go through all 184 of them!!! I had appended the ABC convergence slides to an earlier set of slides on ABC with random forests in case of questions about that aspect, although I did not plan to go through those slides [and I mostly covered the 64 other slides] As the talk was in fine more about an inference method than a genuine Monte Carlo technique, plus involved random forests that sounded unfamiliar to many, I did not get many questions from the audience but had several deep discussions with people after the talk. Incidentally, we have just reposted our paper on ABC estimation via random forests, updated the abcrf R package, and submitted it to Peer Community in Evolutionary Biology!