Turning words into vectors whose geometry means something -- and the tokenizer underneath.
9h · 12 lessons · 1 challenge
Before this: Recurrent Networks & Sequence Models
A model cannot read a word, so something has to turn it into a vector. This module covers both halves: the tokenizer that decides what a unit of text even is, which none of the source courses teaches, and the embedding that gives that unit a position in space.
Then what that space is like: analogies, the embedding matrix, word2vec and negative sampling, GloVe, and the biases that come along with the corpus.
Twenty analogies solved by vector arithmetic. Some of them fail -- say why before you are shown.
Worth 400 XP.
