[Python] Benchmarks for python serialization

[asfimport]

It would be great to have a suite of relevant benchmarks for the Python serialization code in ARROW-759. These could be used to guide profiling and performance improvements.

Relevant use cases include:

• dictionaries of large numpy arrays that are used to represent weights of a neural network
• long lists of primitive types like ints, floats or strings
• lists of user defined python objects

Reporter: Philipp Moritz / @pcmoritz

_{Note: This issue was originally created as ARROW-1391. Please see the migration documentation for further details.}

[asfimport]

Wes McKinney / @wesm:
We already have an ASV (https://github.com/spacetelescope/asv) setup, you can see some benchmarks here https://github.com/apache/arrow/tree/master/python/benchmarks. With some effort we can get these posted nightly someplace like http://pandas.pydata.org/speed/pandas/. See https://github.com/pandas-dev/pandas/tree/master/asv_bench/benchmarks

[asfimport]

Wes McKinney / @wesm:
We should try to get this rolling in 0.8.0. Can anyone take a look at this?

[asfimport]

Wes McKinney / @wesm:
Any chance we can get our ASV setup fixed up in time for 0.8.0?

[asfimport]

Alex Hagerman / @n0mn0m:
I see recent commits in the repo for the benchmarks. Is this still needed? If so any guidance on where the nightly location might be or how to look into this?

[asfimport]

Wes McKinney / @wesm:
See some recent work on this 0ada875.

[asfimport]

Antoine Pitrou / @pitrou:
PyArrow serialization is deprecated, closing as won't fix.