Code to reproduce the NeurIPS submission : "Learning step sizes for unfolded sparse coding"

Compat

This package has been developed and tested with python3.6. It is therefore not guaranteed to work with earlier versions of python.

Install the repository on your machine

This package can easily be installed using pip, with the following command:

pip install numpy
pip install -e .

This will install the package and all its dependencies, listed in requirements.txt. To test that the installation has been successful, you can install pytest and run the test suite using

pip install pytest
pytest

Reproducing the figures of the paper

Figure 2 - Convergence curves between ISTA, FISTA and OISTA.

python examples/comparison_oracle_ista.py

Figure 4 - Learn a small network and display the learned step as well as the distribution of 1/L_S at each layer on the training set.

python examples/plot_learned_steps.py

Figure 5 - Show that when learning a LISTA network with 40 layers, the last layers look like SLISTA layers, where only the step size differs from ISTA. This verifies numerically theorem 4.4.

python examples/plot_dict_similarity.py

Figure 6 - comparison of ISTA, LISTA, ALISTA and SLISTA on large scale problems. Note that this experiment can take up to 24h to be generated. You can tweak the parameter N_JOB and N_GPU in examples/run_comparison_networks.py to accelerate the computations by parallelization and using GPU. Note that if both are set, you will have N_JOB / N_GPU jobs running on each GPU, which should be chosen reasonably.

python examples/run_comparison_networks.py
python examples/plot_comparison_networks.py

Reproducing figures from the appendix

Figure E.1 - Plot the distribution of L_S / L as a function of the cardinal of S.

python examples/plot_sparsity_distribution.py

Figure E.2 - Plot histograms of number of iterations for ISTA, FISTA and OISTA.

python examples/run_comparison_iterative.py
python examples/plot_comparison_iterative.py

Figure E.3 - Plot the test loss relatively to the number of training points used to optimized LISTA or SLISTA.

python examples/plot_learning_curves.py