BERF - A Neural Network for Machine Learning

bert is a neural network that takes as input any arbitrary length of text. It's built on a kind of a sequential model, but what's nice about it is its output is a fixed length vector. So depending on which version of it you choose, you get like either 768 or a hundred and 28 of a vector. And the very surprising thing, or somewhat surprising thing, is that those numbers are a really good way to do automated feature engineering.