Created by: puleon
- new classes structure for siamese networks to perform ranking and paraphrase identification
- new preprocessor class to work with data samples containing few strings for siamese networks
- all components (including tokenizer, vocabularies and embedder) are standard and set in config
- new model with BiLSTM, GRU and max pooling to take into account multi-turn dialogue context
- configs for ranking on ubuntu v2 dialogue corpus with and without taking into account multi-turn context
- new predictor class for siamese networks to work in interact mode
- separate configs for interact mode
- bug fix in SimpleVocabulary class
- tests for all new configs