Some python code for computing word vectors from scratch on Gibbon's Decline and Fall.
model.py: defines the skip-gram model, the loss function and its gradients
data.py: code for convering the HTML book into a token stream
tokens.py: code for building training contexts from the tokens
training.py: a simple SGD training routine
training.ipynb holds an example of training code
For further discussion, see here