Parquet is fine and all, but I love the simplicity and simple interoperability of CSV. You can save a huge amount of overhead just by base64 encoding the vecto

Parquet is fine and all, but I love the simplicity and simple interoperability of CSV. You can save a huge amount of overhead just by base64 encoding the vectors, they aren't exactly human readable anyway. I imagine the resulting file would only be approximately 33% larger than the pickle version.