Using Categorify metadata from full product dataset with transaction subset, is this safe?

[AkramHEMA]

I'm using NVTabular and applying Categorify to categorical features like this:

categ_feats = ['column1', 'column2'] >> nvt.ops.Categorify()

This generates metadata files such as:

1. categories/unique.column1.parquet
2. categories/metadata.column1.parquet

By default, these files are created from the dataset passed into the NVTabular workflow, typically a transactions or interactions dataset, which only includes items that were sold or clicked.

However, I would like to generate these category metadata files from a full product dataset, which includes all possible item_id and category values, not just those that appear in transactions. Then I want to reuse the same categories/ folder when processing the transactions dataset, to ensure consistent and complete category encoding.

My question is:
Can I safely use the categories/ metadata generated from the full product dataset when applying the workflow to a smaller dataset like transactions?

Will NVTabular handle this correctly, without errors or inconsistencies during transformation?

The goal is to define a fixed category mapping based on the full catalog so it remains stable across training, inference, and future data, even if some category values are not present in the current dataset.

Thanks!