figure out encodings for columns

[WeihaoGe1009]

Not all columns needs to be encoded by llm tokenizers.

author: definitely encoder or index
time: no need to put in the model for the current stage, but can do some descriptive stats
title, selftext: input values, llm tokenizer
note: intermediate value? llm tokenizer
jurisdictions: MultiLabelBinarizer
relevance: one-hot
poster's legal status: one-hot
predictions (misconception and unclear knowledge): llm tokenizer or MultiLabelBinarizer?
category: one-hot
background: llm tokenizer or one-hot or MultiLabelBinarizer?

how about prompting? tokenize everything with llm? or other ways to predict + decode?