Keras implementation based on the nano-pelican code in https://github.com/abogatskiy/PELICAN-nano. The reason for rewriting is the possibility of using qkeras for QAT.
It is set up like a python package, so can be pip installed (run command inside folder containing setup.py)
$ C:\users\PELICAN\ pip install .
The folder scripts has code that probably wont be useful for anyone else using this package. The rest of the codebase, however, has been implemented with modularity in mind.
In each subfolder of Experiments, you will find a file called model. These (many different models) may be useful as examples.
All the pelican-related code is implemented as layers. There is quite a few of them
The beginning of any Pelican model. The behaviour of this layer can be adjusted. See here for a complete list of all parameters
Input: (Samples x particles x 4 (3)), where the last dimension is the fourvector of the particle in (E, px, py, pz) or (pt, eta, phi).
Output: (Samples x particles x particles), matrix with all the inner products. Example command:
The behaviour of this layer can be adjusted. See here for a complete list of all parameters
Input: (N x N x L), where each N by N matrix is assumed to be permutation equivariant.
Output: (N x N x 15L), for each input matrix, output the 15 general linear permutation equivariant matrices that can be made from it.
If the input has some extra symmetry, some of the 15 outputs may be redundant. Set
hollow=True if the input matrices have zero trace and symmetric=True if the input matrix is symmetric.
Example command:
$ python train --config=model.yml
See model.yml for example of a config file