Update the training for Hlt2CharmHadInclLcpToKmPpPipBDT
Proposal
In 2016, we implemented the Hlt2 line called "Hlt2CharmHadInclLcpToKmPpPipBDT", which is a BBDT based line, and the trained with 2015 Minimal Biased data for background distributions, 2012 MC for Lambda_c from Xicc decays for signal distributions. Now Use the latest samples for the training. For background distributions now we use 2016 Minimal Biased data, with a smaller sample size, but more similar properties as the 2018 data taking. For signal distributions now we use 2016 MC.
Another purpose for modifying this turbo line is that we want it to benefit more analysis. The following changes are made:
-
During the training, we add the \Lambda_c^+ from \Lambda_b^0 decays, as well as prompt \Lambda_c^+ for the signal distributions.
-
Remove the "\Lambda_c^+ DIRA" and "\Lambda_c^+ IPCHI2" variables during the training, because the signal samples that we use have quite difference lifetime on the particles from which \Lambda_c^+ particles decay. What's more, we do not want to include the IPCHI2 variable because it will complicate the lifetime analysis, especially when it is an input variable of MVA (hard to study its effects).
Latest presentations in trigger meeting
Many thanks to Ao who involved this trigger line in his presentation https://indico.cern.ch/event/692137/contributions/2936672/attachments/1618704/2573987/AoXu_MVA-based-Hlt2-lines.pdf. During the meeting, the rates and efficiencies of several working points are shown. The rate varies from 40 to 60 Hz, and we need to find a balance between rate and efficiency.
Tasks WIP
Now we are waiting for the ParamFiles to be released. Before that we will not be able to have these tunings validated in nightly test.
Another important issure is about whether we should save the full information of the selected events, go to the PersistReco line or a selective PersistReco. From the email I find that trigger experts are discussing on this issue, and the decision is not made yet.