Draft: Fixed GAN Calo Example
This MR adds the first implementation of GANs used in fast simulation and is mostly dedicated to benchmarking the infrastructure. We create here a very simple GAN with PyTorch where we fix all the elements in the tensor in the output in order to have deterministic output from the neural network when testing fast simulations.
Example of the output from neural network as seen by Gaussino:
ToolSvc.FastCal...SUCCESS [ Worker #0 ] Elements of the GAN grid:
| 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, |
| 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, |
| 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, |
| 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, |
| 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, |
| 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, |
| 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, |
| 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, 0.015625, |An example & tutorial can be accessed here https://gitlab.cern.ch/mimazure/gausstrainingsample/-/blob/fixed-GAN-CALO-example-in-Gaussino/fixed_gan_model_description.ipynb
Examples
In the following events output from the notebook given above is given. The neural network generates a grid of 8 x 8 fast hits, centered at the point where the particle passes through the calorimeter.
