Ben Letham

@justdoit0823 @javabrett that certainly does seem like it'd be a good solution, but `setup_requires` is not recommended for use with pip because packages specified in `setup_requires` will be installed using...

I think pinning the version is a bit problematic too - for instance later versions of Stan require C++14. If you have an older OS, you might not be able...

Yeah ultimately in constrained EI the constraint shows up in the acquisition function as Probability(x is feasible). In the usual case where constraints are continuous valued and observed this probability...

The issue we ran into when trying to do something very similar to this was that a sharp transition from p=0 to p=1 (always fails to always succeeds) forced the...

One might gain benefit from including a time series as an extra regressor if 1) that time series might be correlated with the one of interest, and 2) you expect...

That's great. My impression is that a significant part of the use-case for #307 (trends that saturate on only one end) is for time series that are known to always...

Yeah, I don't think it's going to be possible to implement a new trend or likelihood without making changes to both stan and Py/R. You make a good point about...

I just added a first take at negative binomial likelihood in the nb_likelihood branch, with a diff and example notebook in #1544. It seems to be doing something reasonable but...

This is a great example, thanks for coming up with it! I made a tweak to account for the different shape of daily seasonality for working days vs. non working...

@oren0e thanks for sharing this, yeah the numerical issues of the hinge function are pretty significant on this problem. I've been doing some testing of various approaches, and in #1668...