There is a really good explanation of this here:
https://forums.fast.ai/t/to-label-encode-or-one-hot-encode/6057