First study comparing EPD multiplicity (54GeV) to TPC RefMult to try to get centrality using a neural network.

(unnamed) version 0 was presented Jan 15 2018 in the EPD group meeting.

V2 fixes an issue with the calculation of rho on slide 16, adds a slide 17 to explain why the results look this way, and adds a slide 19 comparing EPD multiplicity in EPD centrality bin x to TPC multiplicity in TPC centrality bin x -- a plot Mike requested during the meeting.

V3 includes a plot Mike wanted that shows TPC multiplicity given an EPD/NN centrality bin overlayed on top of the whole TPC multiplicity distribution

V4 is a new NN where I got rid of a cut on the minimum ADC sum for the east and west. This cut should be fine, but if I take it away the 1D multiplicity distributions looks more like a typical RefMult plot (see backup slides in say this version and V3). I don't know why this is, we should have large BBC ADCs in the 54GeV data, but maybe the BBC isn't calibrated in this pre-production (?). The hope is that requiring non-zero adcs we can get rid of the beam pipe events, which I am suspicious of in a pre-production. One certainly sees something weird along the zero ADC/zero TPC mult plots in slide 14.