@nikhilwoodruff feel free to suggest otherwise given these are large files (larger than others), but to be consistent with other generate functions I think we'll want to avoid writing the source files to disk, and instead load from the zip file directly. This might make most sense as a function, something like this (not sure if it'll work):

def concat_zipped_csvs(url: str, prefix: str) -> pd.DataFrame:
    # Creates a DataFrame with the two csvs inside a zip file from a URL.
    zf = ZipFile(BytesIO(requests.get(url)))
    a = pd.read_csv(zf.open(prefix + "a.csv"))
    b = pd.read_csv(zf.open(prefix + "b.csv"))
    res = pd.concat([a, b]).fillna(0)
    res.columns = res.columns.str.lower()
    return res

Then called as:

person_df = concat_zipped_csvs(
    f"https://www2.census.gov/programs-surveys/acs/data/pums/{year}/1-Year/csv_pus.zip",
    "psam_pus"
)

And similarly for household.

I think that's fine as a general approach for large files - though I did envisage something similar to openfisca-uk-data's download() function (either from gcp or GitHub) being useful here

add a newline