Use JMH to benchmark sampling from Gaussian (Normal distribution) vs. EmpiricalDistribution

[cristina-abad]

In our programs, we are sampling from EmpiricalDistribution (http://commons.apache.org/proper/commons-math/apidocs/org/apache/commons/math3/random/EmpiricalDistribution.html). However, our code is currently very slow. One possible reason is that sampling from this distribution may be very inefficient. For this reason, we want to benchmark how long it takes to sample:

1. 1000000 (one million) numbers using Java's default random generator
2. 1000000 numbers from the Normal distribution (http://commons.apache.org/proper/commons-math/apidocs/org/apache/commons/math3/distribution/NormalDistribution.html)
3. 1000000 numbers from the Exponential distribution (http://commons.apache.org/proper/commons-math/apidocs/org/apache/commons/math3/distribution/ExponentialDistribution.html)
4. 1000000 samples from an EmpiricalDistribution with very few bins
5. 1000000 samples from an EmpiricalDistribution with a large number of bins.

Victor: Ask Edwin Boza for two EmpiricalDistributions (one small and one large)

To close this Issue, I will need:

• The commit number of the benchmarking code (including a modification to the Makefile) that has the code to run.
• Code must pass a code review. This means it should be properly indented, it should use appropriate variable names. It should have adequate comments (in English).
• Must include a PDF with a summary of the findings, with tables and/or graphs that allow us to compare the performance of the 5 approaches listed above.

[cristina-abad]

Otros dos tutoriales: http://www.rationaljava.com/2015/02/jmh-how-to-setup-and-run-jmh-benchmark.html http://java-performance.info/jmh/

Y un buen ejemplo: http://stackoverflow.com/questions/81346/most-efficient-way-to-increment-a-map-value-in-java