I got confirmation from Luc.
He also added it to the errata file---the entries regarding p. 511, page
6 of http://luc.devroye.org/errors.pdf
On 12/14/2017 11:11 AM, mp...@tiscali.it wrote:
If Luc's explicit green light will not arrive before it is decision
time, Paolo's point 2- below is doable.
Il 13.12.2017 12:51 Jonathan Wakely ha scritto:
On 12/12/17 21:37 +0100, Paolo Carlini wrote:
Hi, On 12/12/2017 19:42, Michele Pezzutti wrote:
Hi. Yes, I looked at the text before submitting the patch. I
contacted Devroye and he confirmed that another reader had also
pointed out this bug but not the solution. I sent him my proposed
patch, he will look into it (no idea when though).
Nice.
I would state that "comparison function for x = 1 is e^(1/78)"
(which becomes 1/78 as the algorithm uses log-probabilities). I
think the change is needed because otherwise, for that particular
bin, the rejection probability is lower than it should be,
resulting in a higher number of samples.
Ok. Ideally I would be much less nervous about committing the patch
if we either 1- Had Luc's explicit green light; 2- Were able to
*rigorously deduce* within the framework of the book why the change
is needed. That said, the patch makes sense to me and so far holds
up well in all my tests (I'm currently running a full make check). I
would say, let's wait a week or so and then make the final decision.
Jon, do you agree? Ideas about further testing? (eg, some code you
are aware of stressing Poisson?)
No, I have nothing useful to add here, but I CC'd Ed on the PR as I'd
like his input.