Creating input data from simulation

The input data is generated using the scripts and macros in the alice-tpc-offline code repository (link). The code and data are documented in the corresponding JIRA ticket ATO-518. The code is run on the GSI batch farm.

The following steps are performed to obtain the input data used for training and validation:

1. MC simulation of Pb-Pb or pp events, requiring storage of TPC hits
   • Central productions were performed for ML studies. Part of the data was copied to GSI
     • 100k Pb-Pb events JIRA ticket O2-1661
     • 2M pp events, JIRA ticket O2-2803
2. Choosing randomly events from the simulation pool to obtain position and charge of TPC hits
   • N events are chosen to achieve the desired number of ion pile-up events
   • Events are rotated in phi
   • Hits and charge are converted into positive ions
   • Random ion drift time is chosen for position of primary ions and IBF
   • Detailed procedure described in this presentation
3. Obtain n (e.g. n=1000) space-charge density and 3D, 1D IDC maps with random density saved as TH3 histograms (density) and CalPads and vectors (3D, 1D IDCs)
4. Obtain mean space-charge density and 3D, 1D IDC map averaging over all random maps
5. Eventually, scale nominal mean density map by +-3%, +-6% (2 sigma, 4 sigma of the expected relative density fluctuations) to obtain larger distortion fluctuation values when pairing a random with a mean map
6. Calculation of potentials, E-fields, distortions and corrections for all random and mean maps from the density histograms
   • Calculation done with O2 space-charge framework
7. Storing final calculations as TTrees in root files (several maps in one root file)
8. For ML studies, data from TTrees is exported into numpy arrays
   • One numpy file per variable per map
9. Eventually, full map data from numpy files is downsampled (in number of TPC points) and stored as TTrees in root files (called cached training data in the framework)