TMultiLayerPerceptron with arrays

ROOT
#1

I’m calling the following:

TTree* tree = new TTree("tree", "tree);
tree->ReadFile("posn.txt");
TMultiLayerPerceptron* mlp = new TMultiLayerPerceptron("pmt[61]:5,4:x,y", training_tree);

where posn.txt looks like:

pmt[61]/D:x/D:y/D

However, this does not work. When the MLP is created I get the error:

Error in <TTreeFormula::TTreeFormula>: Index 61 for dimension #1 in pmt[61] is too high (max is 60)

It works if I change "pmt[61]:5,4:x,y" to "pmt[60]:5,4:x,y". Is there way to extend this to 61 entries? Also, is this the proper way to use arrays with a TTree/MLP combination? I never found an explicit example of the form it should take but found a few references to its ability. I’m worried the attached image shows that there is only one input and I will need to create an artificially 61 node hidden layer to emulate having 61 inputs. Thanks.

#2

[quote] Is there way to extend this to 61 entries? [/quote]It is already the case (when you use pmt[60]) since pmt[61] leads to TMultiLayerPerceptron to request an out-of-bound element. This is likely a actually a but in TMultiLayerPerceptron, that you could report in savannah.

Cheers,
Philippe.