Flag training

Network finds a correlation, but not very convincing yet

2D Conv net performs significantly better

Sim data

Questions:
- When running the full reco workflow dumped in reco_NOGPU.log I get more tracks than when running with standalone o2-tpc-reco-workflow (approx. 7-8%)
- I currently run with the options from the reco_NOGPU.log, in particular for SC distortions:
  - --lumi-type 1
  - For real data I use (confirmed by Alex):
    
    --lumi-type 2 --corrmap-lumi-mode 1
- I found this statement in one of the MC scripts:
  
  # Setup the TPC correction scaling options for reco; They come from the anchoring setup # Some useful comments from Ruben: # - lumi-type == 0 means no-scaling of corrections with any measure of the lumi rather than no corrections at all. # - The "no corrections" mode is imposed by the TPCCorrMap.lumiMean configurable being negative, in this case all other options in the corrections treatment are ignored. # - But if the MC simulation was done with distortions, then the reco needs --lumy-type 1 (i.e. scale with the CTP lumi) even if the corresponding anchor run reco was using --lumy-type 2 # (i.e. scaling according to the TPC IDC, which don't exist in the MC).
- My question: The real data reco has a significantly different chi2/(2*NCL - 5) distribution than on simulated data with SC distortions
- All my results concerning tracking efficiencies etc. will be incorrect / worse compared to real data
- Is this a problem?
New strategy. Run over digits produced with central MC productions
- alien_find /alice/sim/2025/LHC25b8a_v8b3c/0/559781 debug_full_archive.tgz
- Current problem in reco: IR 49149/0 is ahead of the reference IR 0/47553440

o2-tpc-reco-workflow --session 9 --severity info --shm-segment-id 9 --shm-segment-size 90000000000 --timeframes-rate-limit-ipcid 9 --input-type digits --tpc-digit-reader "--infile $SIMDIR/tpcdigits.root" --output-type tracks --lumi-type 1 --corrmap-lumi-mode 2 --resources-monitoring 2 --configKeyValues "GPU_QA.output=histograms.root;;GPU_proc.runQA=1;GPU_global.rundEdx=1;keyval.input_dir=$SIMDIR;keyval.output_dir=/lustre/alice/users/csonnab/PhD/jobs/clusterization/QA/output/12082025_monalisa_25b8a_v8bc3/cfreg_30/199/reco;GPU_global.deviceType=CPU;GPU_proc.debugLevel=0;GPU_global.synchronousProcessing=1;GPU_proc.clearO2OutputFromGPU=1;GPU_proc.ompThreads=56;GPU_proc.deviceNum=-2;GPU_proc.forceHostMemoryPoolSize=110000000000;GPU_global.overrideNHbfPerTF=128"

Thesis writing

MC chapter almost ready for first reading: Expect to hand it to David end of next week, afterwards to Silvia.