dataset demo: 20newsgroups

[jaberg]

Produce an IPython notebook that illustrates how to train a model on 20newsgroups using the ipython parallelization backend.

Show some training curves, and do some basic analysis of the space to determine what kinds of models worked.

• what preproc do we need
• what version of task to run?

N.B. 20 newsgroups data should not include the "publication details" at the bottom of each file. There's a note to this effect in the sklearn website describing 20newsgroups data set.

[jaberg]

Nevermind the ipython stuff here, let's just make a minimal notebook that illustrates fitting, choosing a pre-processing space, choosing a classifier space, turning the "trials" object into some pretty pictures. All code that we don't critically want to display in the notebook goes into support module that we can easily unit-test. We link to that file from the notebook so that html viewers can look under the hood.