GAN_CP

This repository contains PyTorch implementation of the following paper: Detection of Berezinskii-Kosterlitz-Thouless transition via Generative Adversarial Networks [1]

The code can be used in general for detecting critical points (CP) of physical systems in an unsupervised fashion.

The structure of the code is based on GANomaly: Semi-Supervised Anomaly Detection via Adversarial Training

Installation

1. Clone the repository:

   git clone https://github.com/cerbero94/GAN_CP.git
   

2. Create and activate the virtual environment using conda:

   conda create -n gan_cp python=3.7  
   conda activate gan_cp
   

3. Install the dependencies contained in the requirements file:

   pip install --user --requirement requirements.txt
   

Evaluation of the models in the paper

In order to run the evaluation of the models trained for the paper, execute:

./paper_figures.sh

It will reproduce the plots of Fig. 4 by loading the pre-trained models.

Reference

[1] D. Contessi and E. Ricci and A. Recati and M. Rizzi (2021) "Detection of Berezinskii-Kosterlitz-Thouless transition via Generative Adversarial Networks", arXiv:2110.05383