John Chodera

> In true leave-one-out, you'd use all your counts to train a model, and your test set would be one count I'd be a bit careful here, since this isn't...

For bootstrap or cross-validation sets, what about simply omitting synthetic datasets that don't sample the same state space? This will obviously give you an underestimate that might be severe if...

@maxentile: Can you report more information?

Why drop py27 support sooner?

We're mostly py27. Dropping support for py27 early will hurt us. We are working on migrating everyone, but it's a lot of worth to switching everything. A lot of those...

> I see msmbuilder as a "leaf" in the dependency tree of the scientific python ecosystem. That's basically saying "we don't want you using `msmbuilder` under the hood to build...

> It's hard to get a sense of the community using msmbuilder. There's very little activity on the issue tracker / pull requests / mailing list outside of a couple...

Ardent privacy advocates an always pay attention to the release notes and disable it. I think that is eminently fair for free software that one presumably is electing to install....

We could use post-link scripts for that purpose, I bet: http://conda.pydata.org/docs/building/build-scripts.html

> How would we handle non-conda installs though? A separate prompt during python setup.py install? Can't you just have `setup.py` emit a warning as well? For a warning, something like...