Training and validation part: recycle the energy regression code.
for the input, since z is quantized, add an uncertainty of +/- 0.5 cm (Gaussian) around the true value for the internal points and uniform [-3, +0.5] for z=5 cm and [-0.5,+3] cm for z=48 cm to reduce the border effects
Added the computation part for the friend trees. The z is computed only in the same phase space of the training
The apparent bias in the Z for the intermediate points needs to be checked further (could be overtraining, bad estimate of the true value, bugs...)
Training and validation part: recycle the energy regression code.
Added the computation part for the friend trees. The z is computed only in the same phase space of the training
The apparent bias in the Z for the intermediate points needs to be checked further (could be overtraining, bad estimate of the true value, bugs...)
Some validation plots in this link
Reported here the relevant static plots:
bare output: all together all Zs
bias vs Z
resolution vs Z
weight of input variables: