Note: all python scripts should be run from the root directory of this repository.
In order to start training the model, the data must first be fetched and put in the data
directory.
Afterwards, generate the subsets for training and test by running:
python data/preprocess_data.py
This script will also split each dataset based on their number of the PRI_jet_num
feature.
Inside scripts/models
the following scripts are available:
autoencoder_tf.py
andautoencoder_keras.py
: Instances an autoencoder using the respective deep learning framework. Note that the TensorFlow version was used in this work and this is reflected in the train scripts.variational_autoencoder_tf.py
andvariational_autoencoder_keras.py
. Instances a variational autoencoder.gan.py
. Contains code for instancing the generator and discriminator models of a GAN. These are actually combined with thekeras_adversarial
python package that can be found here on GitHub (it's also a git submodule in this repository).
These models are instanced in their respective training scripts in:
models/autoencoder
models/variational_autoencoder
models/gan
Additional code is available in the scripts/feature_selection
to plot distributions for each feature.