Kyle Beauchamp

System details: anaconda python2.7, pytables 3.1.1, Ubuntu 14.04. I've tried both pymc 2.3.4 and latest git.

I think there is an outstanding pull request already waiting for review. We should see if that fixes the problem. On 4/18/15, John Chodera [email protected] wrote: > @mrshirts: Can you...

We still need someone to merge the pull request. I am not sufficiently familiar with the Gromacs code to review the changes.

I'm OK with that plan--novices should be using conda anyway, and advanced users don't really have a barrier to grabbing the release source tarball.

Do we even need speed here? For example, the SVD algorithm involves an `svd()`, a `pinv()`, and a couple of small matrix multiplies. In my experience, these take under one...

Also, I strongly advise having exactly _one_ implementation, unless someone has a strong and well-documented (in code and literature) justification for several.

I'm still strongly in favor of finding a _single_ method that we can agree on. Reasons below. 1. K = 1000 is a non-issue with linear algebra on modern machines....

My point is that we should, if possible, find a single implementation that works for all cases. I'm fine to support several implementations if that's impossible. Perhaps we should be...

OK, so the next question is for you and Michael to recall some of the "difficult" examples, so we can convert them into regression tests.

I see a 1D US example in pymbar-examples, but no 3D