Andrewnguyen/caliper integrate Caliper, n

[andrewnguyen]

Sort of a big PR:

• Integrates caliper
• Added a second experimental version of HyperLogLogMonoid that does the jRhoW calculation more efficiently (but less idiomatically perhaps?)
• Added tests which confirm that the original HyperLogLogMonoid and the new version produce the same HLL structure using the apply method (perhaps we should be more thorough)
• Added benchmarks comparing the run time of running batchCreate parameterized on the size of the input and on the number of bits used.
• To run the benchmarks do 'sbt algebird-caliper/run'

Example run:

[info]         benchmark bits max        us linear runtime
[info]        WithBitSet    5  10    208.54 =
[info]        WithBitSet    5  20    436.65 =
[info]        WithBitSet    5  30    651.97 =
[info]        WithBitSet   10  10    266.32 =
[info]        WithBitSet   10  20    553.00 =
[info]        WithBitSet   10  30    779.33 =
[info]        WithBitSet   17  10   1751.38 =
[info]        WithBitSet   17  20   3531.91 =
[info]        WithBitSet   17  30   4993.15 =
[info]        WithBitSet   25  10 202083.50 ======
[info]        WithBitSet   25  20 595929.50 ====================
[info]        WithBitSet   25  30 893279.00 ==============================
[info] WithBitSetWrapper    5  10      6.86 =
[info] WithBitSetWrapper    5  20     12.87 =
[info] WithBitSetWrapper    5  30     17.77 =
[info] WithBitSetWrapper   10  10      5.06 =
[info] WithBitSetWrapper   10  20      8.04 =
[info] WithBitSetWrapper   10  30     12.55 =
[info] WithBitSetWrapper   17  10      3.96 =
[info] WithBitSetWrapper   17  20      8.16 =
[info] WithBitSetWrapper   17  30     12.93 =
[info] WithBitSetWrapper   25  10      4.23 =
[info] WithBitSetWrapper   25  20      8.40 =
[info] WithBitSetWrapper   25  30     13.56 =

NOTE: I took a few bits of code from https://github.com/sirthias/scala-benchmarking-template. We should wait on merging in until the licensing situation is clear. Also, it'd be nice if someone can double check I have all the necessary attributions.

[ejconlon]

Scaladoc public classes and methods? Especially "j"...

[andrewnguyen]

"j" shouldn't really be public since it's pretty much internal to the workings of HyperLogLogs but sure.

[johnynek]

add a TODO: we could read a byte at a time while bits >= 8, and that would be faster.

[johnynek]

SUPER EXCITED! Nice work! Way to find a bottleneck and hammer it down!