Chase Geigle

First, thanks for working on this! While this certainly allows for compilation without exceptions enabled, I think the more ideal situation would be to specify an alternate API where the...

Oh fun, this looks like it's actually a bug in ICU! I'll have to test later to see if this is fixed in ICU 59.1 and, if not, we can...

Looks like a problem with ICU. We probably need to bump its version for the next release.

Quickfix for now until I bump the ICU version: ``` # from your build directory sed -i 's/xlocale/locale/' deps/icu-58.2/src/ExternalICU/source/i18n/digitlst.cpp ```

We should consider this formulation (scroll down for the actual paper): http://homepages.inf.ed.ac.uk/s0562315/progs/#pldlm Kenneth even mentions it in his thesis as future work. It looks like it isn't much harder than...

A few different methods from scikit-learn to think about: - Removing features with [low variance](http://scikit-learn.org/stable/modules/generated/sklearn.feature_selection.VarianceThreshold.html#sklearn.feature_selection.VarianceThreshold) - Removing features that perform poorly as [univariate regressors](http://scikit-learn.org/stable/modules/generated/sklearn.feature_selection.f_regression.html#sklearn.feature_selection.f_regression)

You'd need a map from context -> id though, which is probably too expensive. The postings format requires contiguous ids, too, so just saying `id = hash(context)` isn't sufficient.

Sorry for my incredibly slow response... I'm fine with including this in the `contrib/docker` folder. Maybe `contrib/docker/coursera`? Go ahead and issue a PR placing it there and I can merge...

FWIW, the xbacklight backend should still work. I think this is the issue that I was seeing on my machine as well, which motivated PR #44. That PR doesn't fix...