Source code for the paper High-dimensional maximum-entropy phase space tomography using normalizing flows.
Create conda environment:
conda create -n ment-flow python=3.11.5
conda activate ment-flow
Install the mentflow package via pip. This will also install dependencies.
pip install -e .
Install additional dependencies to run experiments:
pip install -e '.[experiments]'
Experiments use hydra. Config files can be found in /experiments/config. Parameters can be overridden with command line arguments. For example:
cd experiments/rec_2d/linear
python train_flow.py device=mps dist.name=swissroll meas.num=7
Results are stored in ./outputs/{script_name}/{timestamp}/ directory created in the working directory. Runtime parameters are stored in ./outputs/{script_name}/{timestamp}/config/.
Several Jupyter notebooks are included to evalate the trained models. To add the conda environment as a jupyter kernel:
pip install ipykernel
python -m ipykernel install --user --name ment-flow
The following command will run all experiments reported in the paper.
cd experiments
./run.sh <device>
My computer uses the "mps" device, so I run ./run.sh mps. Then run the following to make the plots:
cd analysis
./run.sh