Install the library and understand a simple “Hello World!” Bot written in 7 lines of code. Experiment with basic pattern matching rule-based bot.
Learn how to read and prepare data for trainable components.
Build a simple convolutional neural network to solve the named entity recognition task. Master data downloading, preprocessing and batching then train and score the model.
Intro to DeepPavlov configs - a powerfull method to stack models. Study how to train 4 different task-oriented bots on DSTC2 dataset. These include (1) a basic bot, (2) a bot with a database of restaurants, (3) a bot with fasttext embeddings, (4) a bot with attention mechanism over input words.
Implement in DeepPavlov sequence-to-sequence encoder-decoder model with attention mechanism and teacher forcing for chit-chat.