Here are files for SoLID Heavy Gas Cherenkov PID using AI/ML
refer to solid AIML meeting https://solid.jlab.org/cgi-bin/public/ShowDocument?docid=414

about file names
"backgroundno" dir means without background, 
"background3sector" dir means with "3 sector" background as expected in SoLID
"background3sectordouble" dir means with double "3 sector" background
"nodecay" means any pion and kaon decay event are removed and the files are cleaner
"nodecaytrack" means any decay event are removed and central sector is selected by track instead of max Npe
"z*_p*_theta*" are different specific pion/kaon kinematics where phi angle changes, without those means full kinematics
"pmt/quad/pixel" are different readout type
"row" means each row is one event
"1e5" is for number of events

about file format
all in zip csv format
Event recorded in 3 sectors (for solid sim). Each sector has 16 pmt in 4x4 array and each pmt has 4 quad and each quad has 16 pixels. Each event can be readout by npe/pmt/quad/pixel
npe/pmt/quad/pixel files have 1+6+1/48+6+1/192+6+1/3072+6+1 entries per event in each row
  N=1/48/192/3072 numbers of Npe in npe/pmt/quad/pixel for this event
  And 6 numbers of primary track px,py,pz,x,y,z at HGC entrance plane
  And 1 number of label (1 for pion, 0 for kaon)
Files are listed for pion and kaon, for different momentum and angle, for with/without background, and for different readout sensor

note
0. dir "background*_nodecaytrack" should be the data to use, not the others
1. track info is less "important" than npe info, or in another word, used differently
2. z,p,theta,phi from tracking can fully determine kinematics and thus pion Cherekov light location and size
3. How many individual ML modes are needed for HGC PID? not one for full kinematics, not too many for every kinematics. fix z,p,theta and vary phi should be a good compromise.

goal
1. accept as many as pion events and reject as many as kaon events. pion FOM and kaon FOM could take different values, but one value is simpler and 0.9 could be the baseline goal.
2. compare with and without background 
3. compare different kinematics
4. compare different readout type

= internal note ===============================
pion and kaon files are made by solid_gemc of JLAB_VERSION 1.3
scripts are in https://github.com/JeffersonLab/solid_gemc/tree/master/analysis/hgc_study
pion has 0.5 sim safety factor, kaon and background have no sim safety factor

background file "/cache/halla/solid/sim/solid_gemc/SIDIS_He3_JLAB_VERSION_1.3/pass8/farm_solid_SIDIS_He3_moved_BeamOnTarget_0.561e10_skim_HGCwinCF1.root" which has about 9000 good hgc events