[WIP] Benchmarking Daru::Vector statistics methods

[Shekharrajak]

I think we should continue this idea : https://github.com/SciRuby/sciruby/wiki/Google-Summer-of-Code-2017-Ideas#rewrite-slow-parts-of-daru-in-rubex

[Shekharrajak]

PR #483 will fix the travis error

[Shekharrajak]

Comparing benchmark : https://github.com/SciRuby/daru/wiki/Benchmark-comparison-daru-vs-pandas-&-numpy

[v0dro]

Good thought. However, I think we should move to benchmark-ips for the benchmarking: https://github.com/evanphx/benchmark-ips

It provides much more accurate benchmarks and the comparisons will not vary too much between machines.

[v0dro]

Also, can you please use 2 spaces for all indentation in your code?

[Shekharrajak]

Thanks @v0dro for suggestions, I have to refactor it and will use benchmark-ips .

[Shekharrajak]

I have compared various ways (using Array of array, array of vector, Hash of array , Hash of vector) of initializing Daru::DataFrame in this wiki page .

We can see that the fastest way is hash with list and list of Vector takes long time.

For initializing dataframe of 2 rows and 100 columns : 

Using Hash of lists:     8697.3 i/s
Using Hash of Vector:     3501.7 i/s - 2.48x  slower
 Using list of lists:     1228.0 i/s - 7.08x  slower
Using list of Vector:      379.6 i/s - 22.91x  slower