GSI Forum
GSI Helmholtzzentrum für Schwerionenforschung

Home » PANDA » PandaRoot » Event Generators » simpleEvtGen default EvtRandomEngine
simpleEvtGen default EvtRandomEngine [message #9149] Thu, 06 August 2009 17:57 Go to next message
Marius Mertens is currently offline  Marius Mertens
Messages: 55
Registered: January 2009
continuous participant
From: *ikp.kfa-juelich.de
Hi all,

the default EvtRandomEngine used by simpleEvtGen is basically the generator below:
double EvtRandomEngine::random(){
  static unsigned long int next = 1;
  next=next*1103515245+123345;
  unsigned temp=(unsigned)(next/65536) % 32768;
  return ( temp + 1.0 ) / 32769.0;
}

At the first glance, it does seem rather simple, but I don't have sufficient experience with pseudo random number generation to estimate the properties of the sequences generated with above code. Can anybody tell me something about the properties/random quality of the sequences and their suitability for event generation?

Best regards,

Marius
Re: simpleEvtGen default EvtRandomEngine [message #9151 is a reply to message #9149] Thu, 06 August 2009 18:03 Go to previous messageGo to next message
StefanoSpataro is currently offline  StefanoSpataro
Messages: 2736
Registered: June 2005
Location: Torino
first-grade participant

From: *15-87-r.retail.telecomitalia.it
Maybe it could be better to try to contact the original authors of EvtGen (I can give you their mail addresses, or you can find them in the web).
Hiowever, their idea is ho give a simple random generator, and leave then to the experiments to create their proper random generator.

Maybe somebody could comment on what is used in Babar or in the fast simulation framework.
icon4.gif  Re: simpleEvtGen default EvtRandomEngine [message #9161 is a reply to message #9151] Sat, 08 August 2009 12:57 Go to previous messageGo to next message
Marius Mertens is currently offline  Marius Mertens
Messages: 55
Registered: January 2009
continuous participant
From: *dip.t-dialin.net
This is the recommendation I got from David Lange (EvtGen developer and support contact together with Anders Ryd):

Quote:

Our recommendation is to definitely not use EvtRandomEngine for anything production-like. [our documentation I think says this..]


Marius
Re: simpleEvtGen default EvtRandomEngine [message #9162 is a reply to message #9161] Sat, 08 August 2009 13:01 Go to previous messageGo to next message
StefanoSpataro is currently offline  StefanoSpataro
Messages: 2736
Registered: June 2005
Location: Torino
first-grade participant

From: *8-87-r.retail.telecomitalia.it
Exactly what I was saying,
each experiment should write its own random generator.
Re: simpleEvtGen default EvtRandomEngine [message #9163 is a reply to message #9162] Sat, 08 August 2009 15:09 Go to previous messageGo to next message
Marius Mertens is currently offline  Marius Mertens
Messages: 55
Registered: January 2009
continuous participant
From: *dip.t-dialin.net
Then, is there already somebody working on a default random engine which is suitable for production use?
Re: simpleEvtGen default EvtRandomEngine [message #9164 is a reply to message #9163] Sat, 08 August 2009 16:02 Go to previous messageGo to next message
StefanoSpataro is currently offline  StefanoSpataro
Messages: 2736
Registered: June 2005
Location: Torino
first-grade participant

From: *56-82-r.retail.telecomitalia.it
Not inside our colaboration.
I was hoping we could steal something from Babar, Belle or Bes which are also using evtgen...
Thieves and spies are welcome Smile
Re: simpleEvtGen default EvtRandomEngine [message #9165 is a reply to message #9164] Mon, 10 August 2009 15:16 Go to previous messageGo to next message
Bertram Kopf is currently offline  Bertram Kopf
Messages: 110
Registered: March 2006
continuous participant
From: *ep1.ruhr-uni-bochum.de
Hi Marius and Stefano,
as far as I know the default engine in the BaBar-like full simulation software is the "Ranecu" generator. It is also possible to use all other engines provided by CLHEP. Everything is steered via a random control mechanism which makes sure to reproduce sequences when it is needed to re-run specific jobs or even single events from one specific job. I posted already a link to a documentation on the random control package developed by BaBar (see posting:
http://forum.gsi.de/index.php?t=tree&goto=7708&rid=431&S=e6e 3768b7c9b983ec4aae27f79f25413&srch=RandControl#msg_7708
)

Cheers,
Bertram.
Re: simpleEvtGen default EvtRandomEngine [message #9166 is a reply to message #9165] Mon, 10 August 2009 15:48 Go to previous messageGo to next message
Jens Sören Lange is currently offline  Jens Sören Lange
Messages: 193
Registered: June 2005
first-grade participant
From: *physik.uni-giessen.de
Hi Marius et al, if I am correct, this is actually the standard ANSI C random generator (just written out in numbers), as originally proposed by Kerninghan/Ritchie. The period is 2^31 and it is even used for openssl encryption, so I think it is actually not so bad. cheers, Soeren
Re: simpleEvtGen default EvtRandomEngine [message #9167 is a reply to message #9165] Mon, 10 August 2009 16:20 Go to previous messageGo to next message
Marius Mertens is currently offline  Marius Mertens
Messages: 55
Registered: January 2009
continuous participant
From: *ikp.kfa-juelich.de
Hi Bertram,

thanks for your suggestion and the link!
Was this also "simply" plugged into EvtGen for BaBar? I understood from both your description and the documentation you included that the Ranecu based classes are rather a complete random sequence generation framework than just a simple generator?
I'm asking because for the way I use EvtGen up to now just a simple generator would perfectly suffice, granted its properties are considered sufficient for production use.

Maybe it's time for a Panda random task force Wink
Re: simpleEvtGen default EvtRandomEngine [message #9169 is a reply to message #9167] Mon, 10 August 2009 18:02 Go to previous messageGo to next message
Bertram Kopf is currently offline  Bertram Kopf
Messages: 110
Registered: March 2006
continuous participant
From: *ep1.ruhr-uni-bochum.de
Hi Marius,

EvtGen as well as all other event generators are internal packages in the BaBar-like software. An abstract interface makes sure that these generators can be treated in the same way and don't contain any BaBar-related code.

Cheers,
Bertram.
Re: simpleEvtGen default EvtRandomEngine [message #9170 is a reply to message #9166] Mon, 10 August 2009 18:05 Go to previous messageGo to next message
Marius Mertens is currently offline  Marius Mertens
Messages: 55
Registered: January 2009
continuous participant
From: *ikp.kfa-juelich.de
Hi Soeren,

yes you are correct, it's basically ANSI C rand() with a different b parameter.
Actually I didn't want to say the generator is bad in itself, rather that I'm not sure if the random sequences generated by it are suitable for our purpose.
For safety reasons, I'd feel better though if for production use we'd follow both Stefano's and the EvtGen authors' recommendation to use a more advanced generator.
Re: simpleEvtGen default EvtRandomEngine [message #9171 is a reply to message #9169] Mon, 10 August 2009 18:32 Go to previous messageGo to next message
Marius Mertens is currently offline  Marius Mertens
Messages: 55
Registered: January 2009
continuous participant
From: *ikp.kfa-juelich.de
Hi Bertram,

sorry, my question was a bit unclear apparently, in fact I meant something much more trivial:

As the EvtGen EvtRandom interface is extremely simple (basically just the random() function which returns a random double) I was wondering if the more powerful Ranecu based framework including random number management is as easy to be understood and plugged into this single interface class as (just as a random example) one of the PRNGs from Boost?
My motivation here is the learning curve. Getting random numbers from a boost library function is very easy (and would be sufficient for how I invoke EvtGen) but maybe at a later stage the management functions might be helpful?
Re: simpleEvtGen default EvtRandomEngine [message #9172 is a reply to message #9171] Mon, 10 August 2009 21:40 Go to previous messageGo to next message
Bertram Kopf is currently offline  Bertram Kopf
Messages: 110
Registered: March 2006
continuous participant
From: *pools.arcor-ip.net
Hi Marius,
mertens wrote on Mon, 10 August 2009 18:32


As the EvtGen EvtRandom interface is extremely simple (basically just the random() function which returns a random double) I was wondering if the more powerful Ranecu based framework including random number management is as easy to be understood and plugged into this single interface class as (just as a random example) one of the PRNGs from Boost?
My motivation here is the learning curve. Getting random numbers from a boost library function is very easy (and would be sufficient for how I invoke EvtGen) but maybe at a later stage the management functions might be helpful?


As you can see in the class EvtRandomEngine the function random() is defined as a virtual method. That means that you simple have to write a new class derived from EvtRandomEngine and to overwrite this virtual function. In principle you can specify a specific random engine there. In addition you have to set your new defined engine via the static function "EvtRandom::setRandomEngine(newEngine)". An example can be found in the PandaRoot EvtGen package: the test program "ggEvtGen.cc" makes use of the CLHEP-generator "JamesRandom".

Cheers,
Bertram.

Re: simpleEvtGen default EvtRandomEngine [message #9173 is a reply to message #9172] Mon, 10 August 2009 22:41 Go to previous messageGo to next message
Marius Mertens is currently offline  Marius Mertens
Messages: 55
Registered: January 2009
continuous participant
From: *ikp.kfa-juelich.de
Hi Bertram,

actually even simpler once again, sorry to keep bothering Wink
How to plug different random engines into the EvtRandomEngine interface is perfectly clear.
I don't know much (despite the documentation you referred to) about the Ranecu framework with its management abilities, though.
As Stefano pointed out earlier that it would be advantageous to benefit from e.g., BaBar experience with randomness generation it seems worth a look.

So I think the question basically breaks down to this:
Would you consider it useful to plug that framework into EvtRandomEngine or would you consider it kind of an overkill in that context and rather implement a small, simple engine?

I took a look at JamesRandom which (purely from a point of provided function count) is also fairly simple. My guess from the Ranecu documentation is that in comparison it needs much more setup/options (as in complexity being the price one has to pay for more power)?
As I said, to me it'd be reasonable and perfectly sufficient to use something very simple there (like a Boost PRNG, which is what I'm actually using right now), but maybe the additional features of Ranecu might be useful in the future so the higher complexity in implementing this could pay off later?

As a general note I think it'd be useful to have a reasonable default PRNG implementation which is suitable for production use "as is", otherwise at a later stage other people might run into the same issue again and probably start similar considerations on what might be the "best" choice for that purpose.
Re: simpleEvtGen default EvtRandomEngine [message #9177 is a reply to message #9173] Tue, 11 August 2009 11:31 Go to previous messageGo to next message
Bertram Kopf is currently offline  Bertram Kopf
Messages: 110
Registered: March 2006
continuous participant
From: *ep1.ruhr-uni-bochum.de
Hi Marius,

mertens wrote on Mon, 10 August 2009 22:41


So I think the question basically breaks down to this:
Would you consider it useful to plug that framework into EvtRandomEngine or would you consider it kind of an overkill in that context and rather implement a small, simple engine?



In my point of view this question is very simple to answer. We need a proper random management tool for the event mass production (i.e. for the event generation, simulation, digitization, reconstruction as well as for the analysis). In addition it would be also very helpful for the code development to have the possibility to exactly reproduce single events of one job. Consider the following: If someone reports a bug, let's say the application crashes in event number xyz. Then it would be extremely helpful to reproduce this crash by starting directly at event xyz. Also for this purpose it would be nice to have something more powerful. Therefore I think it is important to have/to develop:

o a proper interface to all event generators

o to have a random control mechanism with the possibility to reproduce just specific events of one job

o (at least) to provide the possibility to switch between different random engines

o to provide an easy to use user interface


Cheers,
Bertram.
Re: simpleEvtGen default EvtRandomEngine [message #9178 is a reply to message #9177] Tue, 11 August 2009 11:50 Go to previous messageGo to next message
Marius Mertens is currently offline  Marius Mertens
Messages: 55
Registered: January 2009
continuous participant
From: *ikp.kfa-juelich.de
Hi Bertram,

thanks a lot for your detailed answer (and your patience, I warned you the actual question was very simple Wink )

I agree with the points you brought up and somehow this brings us back to that a common default solution would be good, as full randomness management is certainly more complex than the poor man's solution with Boost I just did for personal use because there seems to be no canonical way for acquiring randomness in Panda yet.
Re: simpleEvtGen default EvtRandomEngine - PndMTRandomEngine Suggestion [message #9218 is a reply to message #9149] Wed, 19 August 2009 16:26 Go to previous messageGo to next message
Marius Mertens is currently offline  Marius Mertens
Messages: 55
Registered: January 2009
continuous participant
From: *ikp.kfa-juelich.de
Hi all,

I was asked to upload the random engine I'm currently using to have an alternative default until an elaborate solution is deployed.
This one is making use of the Boost libraries. So in case it is not available as a default, it'd be nice to have it added (maybe to the external packages or another appropriate place?).

So if anybody with write access to the corresponding source folders feels like replacing the current default, feel free to use this one Wink

Best regards,

Marius
Re: simpleEvtGen default EvtRandomEngine - PndMTRandomEngine Suggestion [message #9231 is a reply to message #9218] Wed, 26 August 2009 13:49 Go to previous messageGo to next message
StefanoSpataro is currently offline  StefanoSpataro
Messages: 2736
Registered: June 2005
Location: Torino
first-grade participant

From: *0-87-r.retail.telecomitalia.it
Just one question:
once I compile my code with this class, how to use this random generator in evtGen macro?
Re: simpleEvtGen default EvtRandomEngine - PndMTRandomEngine Suggestion [message #9232 is a reply to message #9231] Wed, 26 August 2009 14:00 Go to previous messageGo to next message
Marius Mertens is currently offline  Marius Mertens
Messages: 55
Registered: January 2009
continuous participant
From: *ikp.kfa-juelich.de
Hi Stefano,

basically you just create an instance of it (seed is mandatory) and then pass it to the EvtGen constructor like this:

PndMTRandomEngine myRandomEngine(seed);

//Initialize the generator - read in the decay table and particle properties
EvtGen myGenerator("DECAY.DEC","evt.pdl", &myRandomEngine);

(this is an excerpt from the simpleEvtGen source)
Re: simpleEvtGen default EvtRandomEngine - PndMTRandomEngine Suggestion [message #9235 is a reply to message #9232] Wed, 26 August 2009 15:32 Go to previous messageGo to next message
StefanoSpataro is currently offline  StefanoSpataro
Messages: 2736
Registered: June 2005
Location: Torino
first-grade participant

From: *0-87-r.retail.telecomitalia.it
It seems that in SL4.7 there is already an installed version of bost (or at least it appears in my /usr/include), which creates some conflicts. I have spent some time to modify the config.mk to be able to compile your class, but without anye ffort.
Which boost have you used? And how have you changed your config.mk?
Re: simpleEvtGen default EvtRandomEngine - PndMTRandomEngine Suggestion [message #9237 is a reply to message #9235] Wed, 26 August 2009 16:04 Go to previous messageGo to next message
Marius Mertens is currently offline  Marius Mertens
Messages: 55
Registered: January 2009
continuous participant
From: *ikp.kfa-juelich.de
Hi Stefano,

I did not have to make any changes to the config.mk. When the Pnd* files are placed within EvtGenBase, a
make lib
should automatically compile them so they should be usable for a subsequent
make simple if you use it in simpleEvtGen.

Where does your compile fail? Are there any error messages referring to the Boost inclusion?

My boost version is rather old, 1.31.0. However, the interface should not have changed in such a way that it becomes incompatible with more recent releases.
Re: simpleEvtGen default EvtRandomEngine - PndMTRandomEngine Suggestion [message #9238 is a reply to message #9237] Wed, 26 August 2009 16:14 Go to previous messageGo to next message
StefanoSpataro is currently offline  StefanoSpataro
Messages: 2736
Registered: June 2005
Location: Torino
first-grade participant

From: *0-87-r.retail.telecomitalia.it
this is what I get without changing the config.mk:

Toggle Spoiler
Re: simpleEvtGen default EvtRandomEngine - PndMTRandomEngine Suggestion [message #9239 is a reply to message #9238] Wed, 26 August 2009 16:26 Go to previous messageGo to next message
Marius Mertens is currently offline  Marius Mertens
Messages: 55
Registered: January 2009
continuous participant
From: *ikp.kfa-juelich.de
Hi Stefano,

so at least it does find your Boost libraries without any changes, which is basically good.
However, the error you get was if I remember correctly a known bug in older Boost versions where the copy constructor was accidentally invoked upon instantiation. Actually, the explicit static_cast was supposed to work around that problem. A wild (not verified) guess is that casting to UIntType instead might help.
Re: simpleEvtGen default EvtRandomEngine - PndMTRandomEngine Suggestion [message #9240 is a reply to message #9239] Wed, 26 August 2009 17:22 Go to previous messageGo to next message
Elwin Dijck
Messages: 16
Registered: June 2009
Location: Groningen, The Netherland...
occasional visitor
From: *KVI.nl
Hello all,

Perhaps you found this out already, but it seems that actually the Boost interface has changed: they initialize the Mersenne Twister (which actually needs a few hundred integers as initialization) now in a different way than in older versions. Instead of using a single integer, they want a function/functor that gives a sequence of values to initialize the internal state of the Mersenne Twister.

The Boost documentation says "The seeding from an integer was changed in April 2005 to address a weakness." and refers to here.

Also see how ROOT initializes the Mersenne Twister.

So casting will not help, for the newer Boost versions the argument to seed() must be a function or functor. Simplest is to make a small class that has an operator() that just returns the seed number, though it might be better to use one of the more advanced schemes.

Best regards,
Elwin Dijck
Re: simpleEvtGen default EvtRandomEngine - PndMTRandomEngine Suggestion [message #9241 is a reply to message #9239] Wed, 26 August 2009 17:22 Go to previous messageGo to next message
Marius Mertens is currently offline  Marius Mertens
Messages: 55
Registered: January 2009
continuous participant
From: *ikp.kfa-juelich.de
Hi again, you've probably already run into this, but when I cleaned up the class names, I accidentally deleted the "public" from the class declaration. It must say
class PndMTRandomEngine : public EvtRandomEngine

in PndMTRandomEngine.hh. Sorry!
Re: simpleEvtGen default EvtRandomEngine - PndMTRandomEngine Suggestion [message #9242 is a reply to message #9240] Wed, 26 August 2009 17:29 Go to previous messageGo to next message
Marius Mertens is currently offline  Marius Mertens
Messages: 55
Registered: January 2009
continuous participant
From: *ikp.kfa-juelich.de
Hi Elwin,

thanks a lot, I was not aware of that interface change. I'm indeed using an older version, not sure about Stefano, though.

However, as it turns out to be more complex than expected to interface to Boost in a sufficiently generic way, what about then really changing to Root's TRandom3 as was suggested this morning? Especially since they also use a Mersenne Twister anyway?
Stefano, would that maybe easier for you to integrate?
Re: simpleEvtGen default EvtRandomEngine - PndMTRandomEngine Suggestion [message #9243 is a reply to message #9242] Wed, 26 August 2009 17:32 Go to previous message
StefanoSpataro is currently offline  StefanoSpataro
Messages: 2736
Registered: June 2005
Location: Torino
first-grade participant

From: *13-87-r.retail.telecomitalia.it
I think so.
With boost the problem is that by default under sl you have it installed, then you have conflicts not so easy to solve.
I will try to study the "TRandom" way...
Previous Topic: EvtGen Output Fixed Point Precision
Next Topic: Meaning of fSeed for dpm_gen__ in PndDpmDirect
Goto Forum:
  


Current Time: Thu Mar 28 22:44:02 CET 2024

Total time taken to generate the page: 0.00714 seconds