>>>>> Steven Yen >>>>> on Fri, 9 Oct 2020 05:39:48 +0800 writes:
> Oh Hi Arne, You may recall we visited with this before. I > do not believe the problem is algorithm specific. The > algorithms I use the most often are BFGS and BHHH (or > maxBFGS and maxBHHH). For simple econometric models such > as probit, Tobit, and evening sample selection models, old > and new versions of R work equally well (I write my own > programs and do not use ones from AER or > sampleSekection). For more complicated models the newer R > would converge with not-so-nice gradients while R-3.0.3 > would still do nicely (good gradient). I use numerical > graduent of course. I wonder whether numerical gradient > routine were revised at the time of transition from > R-3.0.3 to newer. As R-core member, particularly interested in numerical accuracy etc, I'm also interested in learning what's going on here. I think we (R core) have never heard of anything numerically deteriorating going from R 3.0.x to R 4.0.x, and now you are claiming that in public, you should really post *reproducible* code giving evidence to your claim. As was mentioned earlier, the difference may not be in R, but rather in the versions of the (non-base R, but "extension") R packages you use; and you were saying earlier you will check that (using the old version of the 'maxLik' package with a newer version of R and vice verso) and tell us about it. Thank you in advance on being careful and rational about such findings. With regards, Martin Maechler ETH Zurich and R core team > Not knowing how different your versions of maxLik are > between, I will try as I said I would, that is, use new > version of maxLik from old R and vice versa, and see what > happens. > Sent from my iPhone Beware: My autocorrect is crazy >> On Oct 9, 2020, at 4:28 AM, Arne Henningsen >> <arne.henning...@gmail.com> wrote: >> >> Hi Steven >> >> Which optimisation algorithms in maxLik work better under >> R-3.0.3 than under the current version of R? >> >> /Arne >> >>> On Thu, 8 Oct 2020 at 21:05, Steven Yen >>> <st...@ntu.edu.tw> wrote: >>> >>> Hmm. You raised an interesting point. Actually I am not >>> having problems with aod per se—-it is just a supporting >>> package I need while using old R. The essential package >>> I need, maxLik, simply works better under R-3.0.3, for >>> reason I do not understand—specifically the numerical >>> gradients of the likelihood function are not evaluated >>> as accurately in newer versions of R in my experience, >>> which is why I continue to use R-3.0.3. Because I use >>> this older version of R, naturally I need to install >>> other supporting packages such as aod and AER. >>> Certainly, I will install the zip file of the older >>> version of maxLik to the latest R and see what >>> happens. Thank you. >>> >>> I will install the new maxLik in old R, and old maxLik >>> in new R, and see what happens. >>> >>> Sent from my iPhone Beware: My autocorrect is crazy >>> >>>>> On Oct 9, 2020, at 2:17 AM, Richard M. Heiberger >>>>> <r...@temple.edu> wrote: >>>> >>>> I wonder if you are perhaps trying to solve the wrong >>>> problem. >>>> >>>> If you like what the older version of the aod package >>>> does, but not the current version, then I think the >>>> solution is to propose an option to the aod maintainer >>>> that would restore your preferred algorithm into the >>>> current version, and then use the current R. >>>> >>>> A less good, but possibly workable, option is to >>>> compile the old version of aod into the current R. > ______________________________________________ > R-help@r-project.org mailing list -- To UNSUBSCRIBE and > more, see https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide > http://www.R-project.org/posting-guide.html and provide > commented, minimal, self-contained, reproducible code. ______________________________________________ R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.