the big query reader casts string columns in the to_pandas call to object, which is bad for memory, Gil contention and performance. We should utilise the same strategy as for read_parquet over in dask to preserve the string dtype when the option is set.
the big query reader casts string columns in the
to_pandascall to object, which is bad for memory, Gil contention and performance. We should utilise the same strategy as forread_parquetover in dask to preserve the string dtype when the option is set.