Output to parquet

[mschulist]

For large datasets, parquet files can be far smaller (I still don't get how the compression is sooo good!) than their csv counterparts. It would be nice to be able to write to a parquet file for this reason.

[sdenton4]

Parquet is columnwise storage, meaning that all of the data from the same column is stored in sequence. This tends to be /much/ more easily compressible than row-wise storage.

CSV is also just a crappy data storage format - storing floating point numbers as a bunch of ascii characters wastes a lot of space.