In order to train deep learning models, run the NC_SELI_Imbalance.ipynb  on Google colab 
(In addition to creating the save_training_Data folder) or run the nc_seli_imbalance.py. 
Default model is a resnet-18 and the code provides two options for running the MNIST or
CIFAR 10 datasets. 
Results will be saved in save_training_Data folder under mnsit and cifar10 for different 
imbalance step ratios. 
For each experiment the following data is saved:
    graphs: a class of data used to keep track of the original NC properties by Papyan derived
        from the original codebase.
    W_list: List of sampled last layer weights.
    mu_c_train_list: List of sampled last layer embedding means for each class.
    B_list: List of last layer bias. (Set to zeros for NC experiments)
    train_list_accuracies: List of per class training accuracies.
    test_list_accuracies: List of per class testing accuracies.

    
