- Maximum entropy models;
- Use of
gensimWord2Veckeraslayers.
- Incorporating new features from
gensim; - Implementation of author-topic model;
- Python 3 compatibility;
- More neural networks;
- More available corpus;
- Generative models;
- Support of seq2seq models;
- Gradual fading-out dependence on
Theano, and lesserkerasbut more fundamentalTensorflow; - Spelling corrections and fuzzy logic;
- Other word-embedding models.