commons-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Gilles <gil...@harfang.homelinux.org>
Subject Re: [Vote] Release Apache Commons RNG v1.2 (RC1)
Date Tue, 04 Dec 2018 22:29:52 GMT
On Tue, 4 Dec 2018 16:33:48 +0000, Alex Herbert wrote:
> On 04/12/2018 15:15, Gilles wrote:
>> On Tue, 4 Dec 2018 14:08:59 +0000, Alex Herbert wrote:
>>> +0 (non-binding)
>>>
>>>
>>> The user guide shows updated Performance and Quality tables 
>>> following
>>> the changes to the core implementation for all tables (verses 
>>> release
>>> 1.1). I note that the times for the 
>>> BoxMullerNormalizedGaussianSampler
>>> have all improved nearly 2-fold which is strange.
>>
>> Agreed.  But I can't say much more.  You are welcome to
>> confirm that either the current or the previous values
>> (or neither) are correct.
>> And we'll update the site accordingly.
>
> OK. I ran a quick trial of SamplersPerformance and it seems that the
> BoxMullerNormalizedGaussianSampler should be slower than the others.
> This matches the old results.
>
> commons-rng/commons-rng-examples/examples-jmh > mvn -P benchmark
> 
> '-Dbenchmark=org.apache.commons.rng.examples.jmh.distribution.(Next*|SamplersPerformance.*Gaussian.*)'
> test
>
> Here are the results with 5 iterations:
>
> RNG 	BoxMuller 	Marsaglia 	Ziggurat
> ISAAC 	0.9507838314 	0.5394285144 	0.2749968235
> JDK 	1.0688682596 	0.6440317765 	0.3243021893
> KISS 	0.9858236618 	0.4700058858 	0.2568217134
> MT 	1.0275424389 	0.5268759359 	0.2690456029
> MT_64 	0.8826816707 	0.4557104478 	0.2296754309
> MWC_256 	0.9507522353 	0.4196383302 	0.2114958789
> SPLIT_MIX_64 	0.8377382444 	0.3617303709 	0.1846275404
> TWO_CMRES 	0.8536155368 	0.387562567 	0.1884355236
> WELL_1024_A 	1.0855815254 	0.5994046678 	0.3088309612
> WELL_19937_A 	1.1350548335 	0.6718304685 	0.3952235947
> WELL_19937_C 	1.1518394142 	0.6985743497 	0.419527361
> WELL_44497_A 	1.1535128658 	0.6958294048 	0.4177399342
> WELL_44497_B 	1.1679061263 	0.7237524243 	0.4451146416
> WELL_512_A 	1.0579604922 	0.5813859004 	0.2975178907
> XOR_SHIFT_1024_S 	0.8405457368 	0.3765978036 	0.2007193048
>
> I'll run the benchmark with more iterations...

I was in fact running Java 10 for the benchmarks: "BoxMuller" is indeed
50% faster on it than on Java 8 and "Marsaglia" 10% slower!

>
> Thanks for the clarifications about the reports being in the modules.
> It is not obvious from the site. Now I've found them they look OK.

So, you'll change your vote?

Regards,
Gilles

>
>> They are next to the files, here:
>> https://dist.apache.org/repos/dist/dev/commons/rng/1.2-RC1/binaries/
>> and here:
>> https://dist.apache.org/repos/dist/dev/commons/rng/1.2-RC1/source/
>>
> Downloaded and checked with gpg and sha512sum. All OK.
>
>>> The 'Project Documentation > Source Code Management' page states to 
>>> use:
>>>
>>> git clone http://git-wip-us.apache.org/repos/asf/commons-rng.git
>>
>> Odd.
>> It works for me (TM). :-)
>
> It now works for me too.
>
>> The command to generate the full site for a multi-module project is
>>   $ mvn clean site site:stage
>
> Now I've found the module reports the local build looks good too.
>
> Alex


---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org


Mime
View raw message