-- TomasCap - 05 Sep 2017

Events

The files that are used for training and verification is in the directory:

• Signal: /data/scratch/userspace/pretz/scrappy-platypus-optimization/datafiles/energy.sweets-dec20.xcd
• Background: /data/scratch/userspace/pretz/scrappy-platypus-optimization/datafiles/energy.dec20.run005481.xcd

The files that are used for making the significance maps is in the directory: /data/archive/hawcroot/data/hawc/reconstructed/hawcprod/v2.02.02/config-38125/reco_xcdf/2017

Cuts

• rec.CxPE40 > 0
• rec.coreFiduScale <= 100 ( It is only applied in training stage)

Training

I've only used a quarter of the total of events for training and other quarter for testing because If I use a half or all events, the jobs were cancelled due to they over passed the memory limit

• I trained the BDT using weights. I only use sweets.TWgt in signal events and (sweets.TWgt) (sweets.SWgt) 2 Pi (1- Cos (delangle)) in BKG events.
• A BDT was trained for each bin.
• The optimal cut is when I found the maximum value of Q factor. The events that are used is the same as BDT was trained.
• The Q factor is defined as TPR / sqrt(FPR); Where TPR is the fraction in percent of signal events that were classified correctly (gamma efficiency) and FPR is the fraction in percent of the background events that were classified as gamma (hadron reject).
• With the quarter for training compute the Q factor using the optimal cut that were found before.

In this case I don't use weight when compute the Q factor. I am modified my code for including the weight.

Crab Map

• January and February is used to make the significance map.
• The Simple Analysis Tools was used.