Most similar existing stage is vocabulary ID creation and assignment / tokenization:
Similar two steps:

1. Map each code string into a integer vocab ID (all in codes.parquet): https://github.com/mmcdermott/MEDS_transforms/blob/main/src/MEDS_transforms/fit_vocabulary_indices.py
2. (alongside other normalization steps) join codes to vocab IDs and convert via the metadata file: https://github.com/mmcdermott/MEDS_transforms/blob/main/src/MEDS_transforms/transforms/normalization.py

Open question: How to download/store/access the ohdsi vocab remapping tables?

I wonder if the pipeline's first step should consider that there might be code/vocab_index in codes.parquet and account for that. If we translate m:m, we will most likely introduce a new word in the vocabulary that will require a new vocabulary index. Is it already assumed that the vocabulary fitting will occur later in the whole pipeline?

And @mmcdermott , could you look at the file structure of what I have added so far to confirm if this complies with the framework?

Thanks for the nudge @prenc ; I will try to take a look later today! Also, yes, we should assume that vocabulary fitting will occur later in the pipeline so we do not need to worry about that at this stage.