Transforming TTree with many variables into a 2D image to use as input for a CNN?

TMVA
1

I have a root file that has 360 variables in the TTree. Each variable outputs a 1 or 0 to represent an on or off pixel (18 variables for each row, with a total of 20 rows). Now I want to transform these into a 2D image to feed into the TMVA_CNN_Classification code. How do I properly do that? The example root file is already in an image format, but I’m not sure how to transform my TTree in such a way. Could I simply load each variable one by one in the correct order, and the code knows that it’s taking it in as an image? Such as:

loader.AddVariable(“var1”,“var1”‘I’)
…
loader.AddVariable(“var360”,“var360”‘I’)

loader.PrepareTrainingAndTestTree(
mycuts,
mycutb,
nTrain_Signal=nTrainSig,
nTrain_Background=nTrainBkg,
SplitMode=“Random”,
SplitSeed=100,
NormMode=“NumEvents”,
V=False,
CalcCorrelations=False,
)

model = Sequential()
model.add(Reshape((20, 18, 1), input_shape=(360,)))
… etc.

Or do I have to do another step in between? Thank you so much.

2

Hi @abayyari,

thank you for the question. Maybe @moneta could help?

Cheers,
Marta

3

Thanks for your suggestion, I think it worked and gave good results based off of my ROC plot, but want to be sure it was input correctly.

ROC
ROC1298×950 198 KB

4

Hi,
Yes you can add teh variables one by one as you are doing or you can add directly an array using DataLoader::AddVariableArray as in this tutorial.

Best regards

Lorenzo

1 Like