David J. Harris

I'm currently using Stan on a remote server, and it seems like various parts of my workflow would be much quicker if everything used compressed files. I'm in the process...

If it's just a copy-paste, then my preference would be to make it an option sooner rather than later. But if it would make the refactor more complicated, then delaying...

Okay, thanks. Looks like Boost would use zlib under the hood, and it's distributed under [its own license](http://www.zlib.net/zlib_license.html). It looks BSD-compatible to my non-lawyer eyes (FSF says it's GPL compatible,...

Thanks for the quick response!

@Ironholds: Thanks Oliver! Before you take a stab at it, I think @bblonder and I should figure something out first (below). @bblonder: I just compared my implementation (#1) versus the...

@bblonder There are at least two differences between your code and mine. 1) About 99% of the `samples` that are fed to the kd-ball-checking algorithm are outside of the SVM's...

@bblonder In my code, we: 1. sample from the hypersphere 2. throw out 99% of the points because they're outside the SVM boundary 3. build the KD tree If I...

Cool. Don't have time at the moment, but the first thing I'd check is the denominator. You'll want it to equal the number of full_samples, I think. You can double-check...

Okay, I think I understand. Thinking about it from the perspective of importance sampling in high dimension helped. Some of the khat values were also on the high side, which...

More access to the loss function would be a big help for my work too. Even if the user doesn't have as much control as suggested in the example above,...