Open mauricelanghinrichs opened 3 years ago
I am not the developer of this package, but I have been using it for the past year or so. The ABCDE method was in v1 of this package but was removed because the posteriors it produces have "bias" (see this discourse post by the author). They mention that this bias can be overcome by running it for additional iterations.
The author added the method back, undocumented, after I pointed out that it performs far better on some of the big disease model problems I work on.
I haven't used it with mixed priors, so I cannot comment on your problem, I think it's probably a misplaced promotion thing. I hope this context was helpful though! If you know more about the differential evolution method and it's implementation then I certainly would find it really helpful if you know about the aforementioned problem.
@pcjentsch thanks for your comment! This bias was only mentioned for earlystop=true
, right? I think default is false and I hope that is fine to use (?). The DE method seems to be also faster for my problem, so I would be quite happy to (safely) continue using it.. :D
Thank you @pcjentsch and thank you @mauricelanghinrichs! As soon as I get some free time I would like to fix the state of this package and make a good release, However if you guys can get to it earlier I would very gladly accept a PR... Very sorry about the issues
Hey!
first, thanks a lot for this package! I was particular happy to see the support for (mixed continuous/) discrete priors, and the
smc
method seems to work well in my initial runs so far. However, I noticed a potential bug when using theABCDE
method (not yet released, current master branch of KissABC v3.0.1). When using a discrete prior the parameters from that prior are (sometimes) non-integer values (in contrast tosmc
).Would be awesome if someone could take a look, as I'm excited to try out also the DE method! The following code produces the behaviour for me (Julia 1.6, macOS, Distributions v0.23.12):
Best