Add option to use tdd file format to train in umami
Summary
This MR introduces the following changes
- Adds a new generator base class that is able to read file from tdd format rescale variables and structure them such that they are suited for training taggers in umami
- Add all the child classes for each model
- For all models add a switch to change to this new generator
- Fixed an issue with the "writer" from the old pre-processing which was saving "weight" of a jet e.g. from sampler as one of the NN's inputs (Therefore if anyone would use the weights for training it will give one discriminative feature to the network which actually likely shouldn`t be there)
- Some simplifications with the data generator in models overall (models now get necessary parameters like data dimensions from data generator itself)
- Attempted optimising speed of the new tddgenerator but it is still lacking somewhat compared to the original generator
Solves following issues:
Conformity
Edited by Ivan Oleksiyuk