Support H2O groupby queries in OmnisciOnRay engine

[ienkovich]

We need to cover aggregation API used in H2O groupby queries. Issue doesn't cover other features required for H2O benchmarks (type cast, data transfer etc.)

Current status:

• Query [CI] Adding travis config #1
• Query Making work with PyPI #2
• Query Adds requirements.txt with ray wheel dependencies #3
• Query Fix read_csv for Pandas 0.23 #4
• Query Initial docs #5
• Query Cleaning up and structuring the imports #6
• Query Making changes to readme #7 - modified to avoid assign
• Query Bumping version for PyPI #8
• Query Make setup.py compatible with pip version 10 #9
• Query Fixing bug in insert #10

[ienkovich]

Current issues with remaining queries:

Query #6 - we need to implement std and median aggregators in OmniSci

Query #8 - we need to use window functions to implement groupby.head method. Unfortunately, OmniSci doesn't preserve rows order in partitions when window function is applied without order by. In our case we have sorted order prior groupby call. So, it is possible to translate, but requires sort_values implementation to become backend specific first.

Query #9 - here we have grouby.apply with lambda. Passed lambda uses unsupported corr aggregate. Also it returns new Series built on cell values extracted from corr result. With would be very challenging to support using lazy computations.

[anmyachev]

@ienkovich is this still relevant?