STALE: See the page in the documentation for full detail.
NEW!!! (July 2016): Added word embedding features (Senna) for demonstration.
With 2K training words, it doesn't improve the quality though...
- Tried Glove and reg_param tuning to no avail.
- Maybe the distributional semantics of words cannot patch the remaining errors.
- Or maybe it'll make a difference on a larger training set.