In this file you will find two directories: "code_local_laptop" and "code_cluster". A main machine (local laptop) and a cluster machine were used for these experiments.

Experiments on the a9a and gisette datasets were run via the cluster, and sido0 experiments were run on a local machine (because the cluster was having issues with scipy during the experiments). 

Within the "code_local_laptop" files, you will find a DRO and NCPL file. These correspond to their respective experiments from our paper. 

The NCPL file is relatively straightforward: there is a file which generates the path data used for the path experiments and a file which generates the data used for the CDF experiments. The plot_code_synthetic file has both codes for generating the plots. It suffices to comment out whichever one one wants to use, they are clearly titled within the file as PATHS for the figures of the paths.

The DRO has alg_SSAGDA and alg_SSAGDA_optimized. The latter one is more optimized and is recommended, the former is there for the sake of completeness. The Jupyter notebooks correspond to running the algorithm we study on each of the three datasets. 

Note that the intention is that a relative directory should be created with each run of the code for either the DRO or NCPL file. It should be noted that sometimes Python has difficulties with directories and one may have to use an absolute path depending on how our code is downloadeded and the data is stored.