This took a while, apologies. So, the way I dealt w/ this is not super optimal but I didn't find an easier way once the events were generated. I ran the same exact jobs personally removing the filter. At which point I have (for each DS) two EVGEN files, one w/ filter and one w/o. From these, I produced TRUTH DxAODs, and ran my analysis code over them to find the SUSY process in each event. Once you do this (and knowing the total efficiency from the original job) you can compute the efficiencies, i.e.
Assume you generated 5k events w/ a filter efficiency of 0.5 that includes to processes: proc_1, proc_2
==> This means before the filter you generated ~ 5k * 1/0.5 = 10k events
==> Let's call the number of events per process after the filer n_proc_1_after, n_proc_2_after
Then generate 10k events w/o any filter
==> Let's call the number of events per process as n_proc_1_before, n_proc_2_before
Then the filter efficiency per process is simply n_proc_i_after/n_proc_i_before
Of course, the trick here is that the events that we use to compute the counts before and after the filter are independent. So there is a bit of a statistical uncertainty to the filter efficiencies one would get. You can minimize this by generating more events, but typically this is not really a bottleneck.
Hi Yu-Ting,
This took a while, apologies. So, the way I dealt w/ this is not super optimal but I didn't find an easier way once the events were generated. I ran the same exact jobs personally removing the filter. At which point I have (for each DS) two EVGEN files, one w/ filter and one w/o. From these, I produced TRUTH DxAODs, and ran my analysis code over them to find the SUSY process in each event. Once you do this (and knowing the total efficiency from the original job) you can compute the efficiencies, i.e.
Of course, the trick here is that the events that we use to compute the counts before and after the filter are independent. So there is a bit of a statistical uncertainty to the filter efficiencies one would get. You can minimize this by generating more events, but typically this is not really a bottleneck.
Best, Serhan