Choose timezone
Your profile timezone:
Cut at (HCAL_E+ECAL_E)/truth_E > 0.66, H/E < 0.4 (loose filters for regression, stronger H/E cut for classification)
Make TriForce do filtering on the fly
Look at lower dropout rate
Switch to early stopping instead of fixed batches
For 1D plots, train multiple times per point and get errors
Look at SeLU instead of ReLU (also batch norm?)
Accuracy vs. energy plots for both small energy range and full energy range