Enable to restrict columns for pandas.read_parquet

[hoffmann]

Problem description

In pandas 0.21 the top level funtion read_parquet() was introduced. Both available engines fastparquet and pyarrow support the specifications of columns to read. If you are only interested in certain columns of a dataframe this reduces the io.

• Fastparquet https://github.com/dask/fastparquet/blob/master/docs/source/quickstart.rst#reading

• PyArrow http://pyarrow-xhochy.readthedocs.io/en/latest/pyarrow.parquet.html#pyarrow.parquet.read_table

It should be also possible to specify the columns in pandas.read_parquet().

[gfyoung]

Sounds good to me!

[jreback]

this is actually quite trivial, we just need to pass kwargs thru on the read. and then you can specify columns=, which we could document as a formal kwarg.

PR's welcome!

[jorisvandenbossche]

More in general, should we pass through **kwargs to the actual engine call?