Polynomial Preconditioner for Compressed Sensing

This repository reproduces the experiments in Polynomial Preconditioners for Regularized Linear Inverse Problems.

Written by Siddharth Srinivasan. Please post an issue on the repository page if there is a problem.

Installation.

Run the following commands in sequence to run the experiments.

1. conda update -n base -c defaults conda
2. make conda
3. conda activate ppcs
4. make pip

Troubleshooting:

1. This repository was tested on an NVIDIA GPU. If running on a system without the same, please remove the following packages from environment.yaml:
   • cudnn
   • nccl
   • cupy
2. Additionally, if not using an NVIDIA GPU, please set devnum = -1 for each of the demo_*.py files.
3. When running make pip, git clone git@github.com:mlazaric/Chebyshev.git will error if GitHub ssh keys are not set. Please replace that line in the Makefile with git clone https://github.com/mlazaric/Chebyshev.git and run make pip again.

Data.

For most experiments, the corresponding data are located in the data folder. For the spiral3d_mrf experiment, please run bash download_data.sh in the data/spiral3d_mrf folder to download the data followed by python3 combine.py.

Run experiments.

• All experiments are in the form demo_*.py.
• An experiment can be performed by running, for example, python3 demo_ccs.py.
• The respective Jupyter notebooks (as in, plot_*.ipynb) can be used to generate images and videos.
• The Jupyter notebooks must be started after running Step 3 above.
• Additionally, LaTeX is required to render equations in the plots.

Uninstall.

To uninstall, run the following commands:

1. conda activate
2. make clean

Packages used:

• SigPy
• SymPy
• Chebyshev

DOI.