Discussion about different results in rate when computing manually and with the efficiencyChecker
This issue is opened after yesterday's discussion in the WP3 meeting.
All scripts that were used in the rates calculations for the topo can be found on the MooreAnalysis branch Topo_Neural_Net together with the commands that are used to launch the script.
For the hlt1-filtered Upgrade Minbias, the script: HltEfficiencyChecker/options/hlt2_2or3_toporate_UpgradeMinbias.py
is used and leads to the following rates:
The notation is the following:
Hlt2TwoBody_9669115942028985Decision means only the TwoBody Topo with an MVA Cut of 0.9669115942028985 is run.
When using a chained Allen approach on a Sim10U1 minbias from the bookkeeping, the script
HltEfficiencyChecker/options/hlt2_2or3_toporate_withAllen.py
is used with the corresponding commands being named in the file. The rates that we get are:
with the same notion as before.
When calculating rates offline "per hand", we (@bldelane) get rates about an order of magnitude lower.
The tuples in used in the training are compiled via ganga and all the corresponding scripts can be found here:
https://gitlab.cern.ch/nschulte/topo_neural_network/-/tree/master/ganga
The topo line-file was amended for this calculation to the following (note the loop over different bins of MVA cuts in the end, which is only for developing purposes and not to be merged)