Re-Implementation of "Operationally meaningful representations of physical systems in neural networks"
This repo is our implementation of the NN architecture proposed in H. Poulsen Nautrup, T. Metger, R. Iten, S. Jerbi, L.M. Trenkwalder, H.Wilming, H.J. Briegel, and R. Renner. "Operationally meaningful representations of physical systems in neural networks" (2020). Note that the original implementation can already be found on github, we implemented this one from scratch and chose different approaches e.g. with regards to the training data generation or the model implementation.
.
├── README.md --> this file
├── comm_agents --> package
│ ├── __init__.py
│ ├── data --> data set related code
│ │ ├── __init__.py
│ │ ├── data_generator.py
│ │ ├── data_handler.py
│ │ ├── optimal_answers.py
│ │ └── reference_experiments.py
│ ├── models --> model implementations and trainng
│ │ ├── __init__.py
│ │ ├── model_multi_enc.py
│ │ ├── model_single_enc.py
│ │ ├── trainig_multi_enc.py
│ │ └── trainig_single_enc.py
│ └── utils.py
├── config.json --> run configurations
├── data --> data files
│ └── training
├── lightning_logs --> model logs (e.g. for tensorboard)
├── models --> model directory
├── notebooks --> for visualizations
│ ├── 01_reference_experiment_visualization.ipynb
│ ├── 02_generated_data_exploration.ipynb
│ ├── 03_single_enc_analysis-1.ipynb
│ └── 04_multi_enc_analysis.ipynb
└── setup.py
Assuming you have a fresh virtual environment wiht python 3.7 or higher, run the following code in the root of this project.
pip install -e .Data set generation (with virt env activated):
python comm_agents/data/data_generator.pyNote that this one runs quite a while and it makes sense to use a beefy machine (many cpus) and specify the number of jobs to run in the config.json (e.g. 48).
Model training:
python comm_agents/models/trainig_single_enc.py