Look for the best variable using MC

An analysis was done in order to seen which variables are the best when all variable in the HAWC data stream were used.

Ranking Using MC

First Previous Next Last Exit

Slide 1 of 8

Choose the best variable

The first plot show the variables that have at least one ranking point with a value less or equal to 2. The second is the same but it has a value less or equal to 4.

Variable that have at least one ranking point with a value less 3

Variable that have at least one ranking point with a value less 5

First Previous Next Last Exit

Slide 2 of 8

New Network Training using Real Data

Three networks were trained using Real Data:

Using 15 inputs:

Using 4 inputs:
- Compactness
- rec.PINC
- rec.LDFAmp
- rec.LDFChi2

Using 6 inputs:
- Compactness
- rec.PINC
- rec.LDFAmp
- rec.LDFChi2
- rec.SFCFChi2
- rec.gammaleLLH

First Previous Next Last Exit

Slide 3 of 8

Conditions

Real data were used to training and verification.
- /data/scratch/userspace/pretz/scrappy-platypus-optimization/

Both stages were applied the following cuts:
- rec.coreFideScale < 100 [Events are inside of the HAWC ].
- rec.CxPE40 > 0.

Crab maps were made using the month: January and February 2017
- /data/archive/hawcroot/data/hawc/reconstructed/hawcprod/v2.02.02/config-38125/reco_xcdf/

First Previous Next Last Exit

Slide 4 of 8

Real data vs MC

In the training stage, I should apply a cut in Energy

Comparative of the NN training with 15 inputs when MC and Real data were used

Comparative of the NN training with 15 inputs when MC and Real data were used -- NN/SC

First Previous Next Last Exit

Slide 5 of 8

Look the maps in Bin 0

Significance maps of Crab nebula using the Networks trained

First Previous Next Last Exit

Slide 6 of 8

Ranking using real data

I don't know why rec.nTankAvail have a good ranking. Maybe it has different value in both files (MC as signal and Real data as Bkg)

Variable that have at least one ranking point with a value less 3 using real data

First Previous Next Last Exit

Slide 7 of 8

Weekly goals

Work on the version 2 of disMax
Read information about BDT
Train a BDT with 15 inputs
Research why don't have a better result when I use real data (Maybe I have a better result if I apply one cut in Energy)

First Previous Next Last Exit

Slide 8 of 8