how to tokenize the trajectory / path?

Thank you for open-sourcing this excellent work — the codebase is really impressive!  
I couldn’t locate the trajectory-tokenization code: could you clarify whether you re-use the pre-trained FAST tokenizer or train a new one from scratch? like `tokenizer.fit(array)`