[gate-users] Random seed

[Dirk Kruecker]

Dear Martin,

I guess the random number generation in GATE can be somewhat
confusing. I better explain everything in detail.

-------------------------------------------------------------
The default behaviour:

Without any 
   /gate/output/root/setSaveRndmFlag n
   /random/resetEngineFrom xxx
command GATE will create 2 random seed files:
   beginOfRun.rndm and endOfRun.rndm.
If you do not change these files the next run will be initialized from the 
endOfRun.rndm file (the beginOfRun.rndm will not change anyhow). So by 
default each run will be different.

The random number engine will be initialized from the endOfRun.rndm.
If there is no  endOfRun.rndm   GATE will create a  beginOfRun.rndm and will 
start from this file. The  beginOfRun.rndm  will _always_  be the same! The   
endOfRun.rndm   will change after each run.

------------------------------------------------------------
Rerunning with the same random numbers:

If you want to re-run the same events you can set 
  /gate/output/root/setSaveRndmFlag 0
GATE will be initialized internally and each run will be the same.

----------------------------------------------------------------
Continuning an existing simulation to increase the number of events:

In the case where you already run a simulation and want to continue to get 
better statistics you need the endOfRun.rndm from your first simulation.
  /gate/output/root/setSaveRndmFlag 1
GATE will automatically look for the endOfRun.rndm file in directory where it 
runs.

--------------------------------------------------------------------------------------
BTW
When the SaveRndmFlag is set you can observe the status of the random 
generator at run end. There are some lines before GATE stops:

----- HepJamesRandom engine status -----
 Initial seed = 9876
 u[] = 0.207714 ...

I skip the 103 numbers  describing the state of the random number engine. If 
(at least 1 of) these 103 numbers are different  your events will be 
statistically independent.

--------------------------------------------------------------------------------------
Continuning an existing simulation to increase the number of events in the 
case where you did not save the endOfRun.rndm from the first simulation:

Here the safest method is to use a seed which is totally different from what 
GATE creates. I have included a zip file with a few seeds to create 
independent random number streams.

  /gate/output/root/setSaveRndmFlag 0
  /random/resetEngineFrom  StartSeq(n).rndm

Just giving the filename without   /gate/output/root/setSaveRndmFlag 0 will 
not be sufficient since GATE will still look for the endOfRun.rndm file.


I hope this helps

Dirk


-------------------------------------------------------------------
Dear Dirk,
 
First we want to get exactly the same set of events in two simulations, the 
detector type #1 and the detector type #2 simulations. (As described in my 
previous email, the first time we attach detector type #1 as crystalSD, and 
detector type #2 as phantomSD, and in the second run - with the same seed - 
we switch the crystalSD and phantomSD for the detectors. This produces two 
singles files which we later sort into sinograms.) This part we have been 
able to accomplish and we are getting good data that we can reconstruct. 
 
After this, however, we want to increase statistics for our simulations, so we 
want to run the same setup/simulation, but this time with a different random 
seed. This is the part which does not work for us. The data that we get 
somehow produces almost identical sinograms as our previous runs, so we do 
not achieve higher statistics.
 
Martin
 
PS. We have also tried setting "/random/resetEngineFrom endOfRun.rndm" as 
suggested by Mutanga Theodore, but with the same results as using 
beginOfRunDummy.rndm.
 
 
-----Dirk Kruecker <d.kruecker at fz-juelich.de> wrote: -----

Dear Martin,

What do you want to achieve? Do you want to get exactly the same
set of events? This is impossible if you change your setup. At some point the 
simulation will differ since it has to simulate the difference in your design 
and it will consume a different amount of random numbers.

?

Best Regards

Dirk

> Hello GATE users,
> We are having trouble with the random seed in GATE. We are developing a PET
> system with two different detectors, and we therefore run GATE twice with
> the same seed. The first time we attach detector type #1 as crystalSD, and
> detector type #2 as phantomSD, and in the second run - with the same seed
> - we switch the crystalSD and phantomSD for the detectors. This produces
> two output (singles) files which we later sort into sinograms. This
> approach is working with the exception of producing original data in
> different runs. If we subtract sinograms from two different runs (with
> different seeds) from each other, it turns out that all events, except for
> a few hundred, are identical. 
> We have tried several different approaches to getting original data. One
> approach is running a short dummy program, and copying the endOfRun.rndm
> file onto our random-file and then running GATE (both times) with :
> /gate/output/root/setSaveRndmFlag 0
> /random/resetEngineFrom beginOfRunDummy.rndm
>
> We have also tried setting "/gate/output/root/setSaveRndmFlag 1" for the
> first run, and copying beginOfRun.rndm onto our random file for the second
> simulation, but that doesn't produce correlated data. Does anybody have an
> idea of how to solve this problem?
> Sincerely,
> Martin Janecek
-------------- next part --------------
A non-text attachment was scrubbed...
Name: seeds.zip
Type: application/x-zip
Size: 229470 bytes
Desc: not available
URL: <http://lists.opengatecollaboration.org/mailman/private/gate-users/attachments/20050630/725bf5be/attachment.bin>