A neural language model, intended to produce embeddings for a linear classifier
Switch branches/tags
Nothing to show
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
scripts
.hgignore
README

README

A neural language model, intended to produce embeddings for a linear classifier

Code initially imported from
    ssh://lgcm.iro.umontreal.ca/repos/language-model.predict-final-word

TODO:
    * sqrt scaling of SGD updates
    * Use normalization of embeddings?
    * How do we initialize embeddings?
    * Use tanh, not softsign?
    * When doing SGD on embeddings, use sqrt scaling of embedding size?