The previous calculation of ML input features accidentally caused the full lepton information to get materialized. This results in a significant I/O increase and slowdown. This update avoids the overtouching by performing the relevant piece of the calculation with a minimal re-built set of four-vectors with all other information stripped out.
This is expected to have significant performance impact for I/O bound setups.
The previous calculation of ML input features accidentally caused the full lepton information to get materialized. This results in a significant I/O increase and slowdown. This update avoids the overtouching by performing the relevant piece of the calculation with a minimal re-built set of four-vectors with all other information stripped out.
This is expected to have significant performance impact for I/O bound setups.
see https://github.com/CoffeaTeam/coffea/issues/892 for details