Source code for the experiments and figures of the paper "Adaptive Acceleration of Sparse Coding via Matrix Factorization ".

• numpy 1.10+
• matplotlib 1.8+
• tensorflow 0.9+
• scikit-learn 1.16+

All the development was done with python3.4 and might not work for earlier versions.

Use the main script NIPS_figures.py to launch the experiements. Various option are available from the command line. See python NIPS_figures.py --help for more information.

To generate the 4 figures from the paper, use:

python NIPS_figures.py --data artificial --save_dir layer1
python NIPS_figures.py --data artificial --rho .2 --save_dir layer2
python NIPS_figures.py --data mnist --lmbd .1 -K 100 --save_dir mnist
python NIPS_figures.py --data images --lmbd .05 --save_dir images