* Seq2Seq goal-oriented bot skill * Ranking model * SQuAD model * 20 classes NER model * Multithreading version of `spacy_tokenizer` for huge datasets * Out of the box REST API for all models * Simplified data handling for csv/json classification dataset * Collection of commonly used layers for TF models * Many bug fixes and minor improvements
0.0.2
* `chainer` for train/inference pipeline building instead of "Mother" model * New abstractions: `Component`, `Estimator`, `NNModel` * New config structure * Bug fixes and minor improvements
0.0.1
* JSON config * "Mother" models * TFModel and Keras model abstract classes * Pre-trained speller * Pre-trained NER / slot filler * Pre-trained goal-oriented chatbot * Pre-trained intent classifier * Pre-trained embeddings for the Russian language * Training and inferring from command line * Telegram UI