 |
 |
Hello Kyle,
I’ve copied the files Pelle referenced to JLab. You can find them at:
/work/hallb/hps/ngraf/MC
Norman
From: [log in to unmask] [mailto:[log in to unmask]] On Behalf Of Kyle McCarty
Sent: Wednesday, June 03, 2015 4:36 PM
To: Hansson Adrian, Per Ola
Cc: Holly Szumila-Vance; Omar Moreno; hps-software; Nathan Baltzell; Graham, Mathew Thomas
Subject: Re: Trigger acceptance in MC
Hello Pelle,
For SLiC, you probably want to talk to Matt or Sho. I think that's more of a Monte Carlo simulation expert question that an Ecal one.
If you have an SLCIO file, I can check it, but I can't really get at it easily on the SLAC system. I've CC'd Matt on this email. Maybe he can help both these? If I can get the base SLiC output file, I can try running readout and see if I reproduce errors and look into where they come from.
- Kyle
On Jun 3, 2015 1:30 PM, "Hansson Adrian, Per Ola" <[log in to unmask]> wrote:
Hi All,
here is the topic that I brought up during the meeting.
Here is a std hep file that you can use if you don’t have one (at SLAC, not sure where to find them at JLab):
/u/br/mgraham/hps/DarkPhoton/SignalEvents/ap1.1gev40mevall_1_200ux40u_beamspot_gammactau_0cm_30mrad.stdhep
If you want to avoid slic and bunching there are files here with that done:
/nfs/slac/g/hps3/users/phansson/
I think the best would be best if an ecal expert can simply start from a std hep file and run slic->bunching->readout sim-> recon on some well known process on the HPS-EngRun-Nominal-v1 detector, like the A’ file above.
I didn’t try Holly’s no-pile up as I would like to have the full treatment of the ecal but it would be good to verify how to run that too but to me personally it has less priority than just having something that works with the standard readout simulation that we are supposed to use unless I know what I’m doing.
The steering files I use are:
EngineeringRun2015TrigPairs1.lcsim
EngineeringRun2015FullReconMC.lcsim
I added a section in this existing page that is supposed to tell us what steering files that is recommended:
I suggest each subsystem take the responsibility of keeping this page (and files) updated.
I hope this helps track down the issue.
Thanks,
Pelle
On Jun 2, 2015, at 11:29 AM, Omar Moreno <[log in to unmask]> wrote:
Hi,
Was the issue with the Ecal energies resolved?
--Omar Moreno
On Tue, May 26, 2015 at 4:35 PM, Kyle McCarty <[log in to unmask]> wrote:
Hello Nathan,
I had one a long time ago, but I don't think it exists anymore. It's kind of a pain to do; you have to insert a bunch of empty events in between each Monte Carlo event (around 150 is safe) and then run the reconstruction like normal, except stop before clustering. Because the FADC hits are not separated by more event than 150, you know that every FADC hit belongs to the whatever the last event containing raw hits was and can collect them all and write them out in a single event.
- Kyle
On Tue, May 26, 2015 at 7:30 PM, Nathan Baltzell <[log in to unmask]> wrote:
Very nice picture explanation!
Does a driver exist in hps-java to take care of this and merge it back into a
format like real data, independently of the clustering algorithm? Shouldn't it?
On May 26, 2015, at 7:02 PM, Kyle McCarty <[log in to unmask]> wrote:
> Hello Holly,
>
> You aren't using running pedestals in MC- it's not even using this driver
>
> Do you mean that it's not used by anything in the driver (and could therefore just be removed entirely) or that it isn't present in the driver? It is definitely in my version of that steering.
>
> Holly, what does your driver actually look at for hits? If it looks at raw hits, you are fine with Monte Carlo. If you look at FADC hits, though, you would need to be able to look across multiple events to form your clusters. Consider the graphic below:
>
> <HitDist.png>>
> The raw hits (the unprocessed SLiC output) is converted into FADC hits, but the FADC hits that correspond to the cluster are actually spread across six events (FADC hits are only written out every other event to simulate a clock-cycle) different events. Thusly, to correctly form this cluster, you would need to retain all six of those events. Otherwise, you are probably forming three different clusters from the same cluster (or possible just losing energy, since the latter hits may below threshold).
>
> - Kyle
>
> On Tue, May 26, 2015 at 6:48 PM, Holly Vance <[log in to unmask]> wrote:
> Hi Pelle,
>
> A few things to check:
> What does the timing of the hits in the cluster look like? There is a time cut set in the ReconClusterer, but it is probably going to be slightly different for MC (not sure).
>
> In general, these problems tend to arise when making the hits into readout hits. (You aren't using running pedestals in MC- it's not even using this driver). I suspect the issue is in EcalRawConverter.
>
> On Tue, May 26, 2015 at 6:32 PM, Hansson Adrian, Per Ola <[log in to unmask]> wrote:
> Thanks Holly,
>
>
>
> On May 26, 2015, at 3:11 PM, Holly Vance <[log in to unmask]> wrote:
>
>> Hi Pelle,
>>
>> I have a simple steering file here that can take in your non-spaced events:
>> org.hps.steering.users.holly.EcalReadoutNoPileUp
>>
>
> Cool. That is good to have.
>
>> This will treat each event independently and can tell you if you are having problems in your readout of the data (it will show you where the peaks are and if you did spacing correctly, etc). This uses the full Recon Clusterer which works fine with Monte Carlo (as least I have never had any problems with it).
>>
>
> Right now I’m trying to use our official simulation (see other email, didn’t see your reply before sending it) which includes spaced out events. The energy of the recon clusters are a little weird. Positions doesn’t seem crazy though. Any ideas what can go wrong?
>
> /pelle
>
>
>> -Holly
>>
>>
>> On Tue, May 26, 2015 at 5:50 PM, Kyle McCarty <[log in to unmask]> wrote:
>> Hello Pelle,
>>
>> Your cluster energy distributions definitely look different from what I saw (see the attached PDF). What clustering algorithm are you using? You need to use the GTPClusterer algorithm for it work right with Monte Carlo; you might be using GTPOnlineClusterer, which only works for the EvIO readout. I'm not sure why you would get more When I run reconstruction for Monte Carlo, I used the following drivers (note that I didn't include tracking):
>> • org.hps.conditions.ConditionsDriver
>> • org.hps.readout.ecal.FADCEcalReadoutDriver
>> • org.hps.recon.ecal.EcalRawConverterDriver
>> • org.hps.recon.ecal.cluster.GTPClusterDriver
>> • org.hps.readout.ecal.FADCPrimaryTriggerDriver
>> You need the GTPClusterDriver to properly handle Monte Carlo hit readout, GTP clustering to make sure that you are simulating the hardware, and then the FADCPrimaryTriggerDriver simulates the hardware pair trigger. Are you using these same drivers?
>>
>> - Kyle
>>
>> On Tue, May 26, 2015 at 5:42 PM, Hansson Adrian, Per Ola <[log in to unmask]> wrote:
>> Hi Again,
>>
>> following up on this topic.
>>
>> Running recon:
>>
>> java -jar target/hps-distribution-3.3.1-SNAPSHOT-bin.jar /org/hps/steering/recon/EngineeringRun2015FullRecon.lcsim -DoutputFile=outfile -i bunched-readout.slcio
>>
>> I see ECal energies that don’t look right. Any idea?
>>
>> I see about 180 tracks in these 1342 triggered events. They look ok but the number seems low?
>>
>> Am I forgetting something again?
>>
>> /Pelle
>>
>>
>> <Screen Shot 2015-05-26 at 2.22.35 PM.png>
>> <Screen Shot 2015-05-26 at 2.31.44 PM.png>
>>
>>
>> On May 26, 2015, at 1:43 PM, Hansson Adrian, Per Ola <[log in to unmask]> wrote:
>>
>>>
>>> Ok,
>>>
>>> using 150 bunches on 10k events I get ~13% acceptance.
>>>
>>> Thanks,
>>> pelle
>>>
>>>
>>> Trigger Processing Results
>>> Single-Cluster Cuts
>>> Total Clusters Processed :: 20295
>>> Passed Seed Energy Cut :: 14119
>>> Passed Hit Count Cut :: 11938
>>> Passed Total Energy Cut :: 10991
>>>
>>> Cluster Pair Cuts
>>> Total Pairs Processed :: 3168
>>> Passed Energy Sum Cut :: 2942
>>> Passed Energy Difference Cut :: 2941
>>> Passed Energy Slope Cut :: 2755
>>> Passed Coplanarity Cut :: 1342
>>>
>>> Trigger Count :: 1342
>>>
>>> Trigger Module Cut Values:
>>> Seed Energy Low :: 0.050
>>> Seed Energy High :: 6.600
>>> Cluster Energy Low :: 0.060
>>> Cluster Energy High :: 0.630
>>> Cluster Hit Count :: 2
>>> Pair Energy Sum Low :: 0.200
>>> Pair Energy Sum High :: 0.860
>>> Pair Energy Difference :: 0.540
>>> Pair Energy Slope :: 0.6
>>> Pair Coplanarity :: 30.0
>>> FADCPrimaryTriggerDriver: Trigger count: 1342
>>>
>>>
>>>
>>> On May 25, 2015, at 3:23 PM, Kyle McCarty <[log in to unmask]> wrote:
>>>
>>>> Hello Pelle,
>>>>
>>>> If you are reconstructing everything from a SLiC file, you need to space out the events because each one contains an A' event and will produce weird pile-up otherwise. I usually insert around 150 empty events between each real event to ensure that the hits and clusters (which are displaced by around 60ish events from the source) do not overlap at all. You can do this with the command:
>>>>
>>>> java -cp $HPS_JAVA org.hps.users.meeg.FilterMCBunches $INPUT $OUTPUT -e150 -a
>>>>
>>>> You should definitely get better acceptance that that as well; I got somewhere between 15% - 20% acceptance for 40 MeV (I don't have the exact value on hand).
>>>>
>>>> Note also that there are (or at least were last I checked) differences in how the Monte Carlo simulation is stored versus the EvIO. In the EvIO data, all the hits and clusters for an event are stored in the same event, but in Monte Carlo readout, they are spaced across several events with each event representing 2 ns of time. As such, some drivers that work for EvIO readout do not for the Monte Carlo (this mainly affects clustering). Sho can correct me if I am misrepresenting something here.
>>>>
>>>> - Kyle
>>>>
>>>> On Mon, May 25, 2015 at 6:08 PM, Hansson Adrian, Per Ola <[log in to unmask]> wrote:
>>>> Just realized, how is the bunch spacing simulated in the readout step? I suppose I need to add some amount of “time” between the events or do a “NoPileUp” simulation...
>>>>
>>>> /Pelle
>>>>
>>>> On May 25, 2015, at 3:04 PM, Hansson Adrian, Per Ola <[log in to unmask]> wrote:
>>>>
>>>>> Hi,
>>>>>
>>>>> Have some issues getting the MC simulation to run.
>>>>>
>>>>> I’m running readout and recon over some 1.1GeV 40MeV A' MC files locally with the trunk. Slic file is here:
>>>>>
>>>>> /nfs/slac/g/hps3/users/phansson/ap1.1gev40mevall_1_200ux40u_beamspot_gammactau_0cm_30mrad_SLIC-v04-00-00_Geant4-v10-00-02_QGSP_BERT_HPS-EngRun2015-Nominal-v1.slcio
>>>>>
>>>>> I’m using Nominal-v1 detector and I simulate readout with:
>>>>>
>>>>> java -jar target/hps-distribution-3.3.1-SNAPSHOT-bin.jar ../kepler2/hps-java/steering-files/src/main/resources/org/hps/steering/readout/EngineeringRun2015TrigPairs1.lcsim -Ddetector=HPS-EngRun2015-Nominal-v1 -Drun=2000
>>>>>
>>>>>
>>>>> I see 245 accepted events out of 10k events which is much lower than I would naively expect?
>>>>>
>>>>>
>>>>> I then run recon with:
>>>>>
>>>>> java -jar target/hps-distribution-3.3.1-SNAPSHOT-bin.jar ../kepler2/hps-java/steering-files/src/main/resources/org/hps/steering/recon/EngineeringRun2015FullRecon.lcsim -DoutputFile=outfiles/ap1.1gev40mevall_1_200ux40u_beamspot_gammactau_0cm_30mrad_SLIC-v04-00-00_Geant4-v10-00-02_QGSP_BERT_HPS-EngRun2015-Nominal-v1-recon -i outfiles/ap1.1gev40mevall_1_200ux40u_beamspot_gammactau_0cm_30mrad_SLIC-v04-00-00_Geant4-v10-00-02_QGSP_BERT_HPS-EngRun2015-Nominal-v1-readout.slcio
>>>>>
>>>>> And I basically get to the 2nd event and then it just hangs there for 5-10mins.
>>>>>
>>>>>
>>>>>
>>>>> Are these the right steps with the latest and greatest sw? Any ideas?
>>>>>
>>>>> /pelle
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>>>>>> Use REPLY-ALL to reply to list
>>>>>
>>>>> To unsubscribe from the HPS-SOFTWARE list, click the following link:
>>>>> https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=HPS-SOFTWARE&A=1
>>>>>
>>>>
>>>>
>>>> Use REPLY-ALL to reply to list
>>>>
>>>> To unsubscribe from the HPS-SOFTWARE list, click the following link:
>>>> https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=HPS-SOFTWARE&A=1
>>>>
>>>>
>>>
>>>
>>> Use REPLY-ALL to reply to list
>>>
>>> To unsubscribe from the HPS-SOFTWARE list, click the following link:
>>> https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=HPS-SOFTWARE&A=1
>>>
>>
>>
>> Use REPLY-ALL to reply to list
>>
>> To unsubscribe from the HPS-SOFTWARE list, click the following link:
>> https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=HPS-SOFTWARE&A=1
>>
>>
>>
>> Use REPLY-ALL to reply to list
>>
>> To unsubscribe from the HPS-SOFTWARE list, click the following link:
>> https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=HPS-SOFTWARE&A=1
>>
>> NOTE: This message was trained as non-spam. If this is wrong, please correct the training as soon as possible.
>> Spam
>> Not spam
>> Forget previous vote
>
> NOTE: This message was trained as non-spam. If this is wrong, please correct the training as soon as possible.
> Spam
> Not spam
> Forget previous vote
>> Use REPLY-ALL to reply to list
>
> To unsubscribe from the HPS-SOFTWARE list, click the following link:
> https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=HPS-SOFTWARE&A=1
>
>
>
>
> Use REPLY-ALL to reply to list
>
> To unsubscribe from the HPS-SOFTWARE list, click the following link:
> https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=HPS-SOFTWARE&A=1
>
Use REPLY-ALL to reply to list
To unsubscribe from the HPS-SOFTWARE list, click the following link:
https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=HPS-SOFTWARE&A=1
Use REPLY-ALL to reply to list
To unsubscribe from the HPS-SOFTWARE list, click the following link:
https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=HPS-SOFTWARE&A=1
NOTE: This message was trained as non-spam. If this is wrong, please correct the training as soon as possible.
Spam
Not spam
Forget previous vote
Use REPLY-ALL to reply to list
To unsubscribe from the HPS-SOFTWARE list, click the following link:
https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=HPS-SOFTWARE&A=1
Use REPLY-ALL to reply to list
To unsubscribe from the HPS-SOFTWARE list, click the following link:
https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=HPS-SOFTWARE&A=1