There are two experiments:
    1. Polynomial effective class expressivity (expressivity_experiment.py)
    2. Maximum accuracy in the strategic setting (approximation_experiment.py)

Note that both experiments are designed to run over multiple CPUs to parallelize
calculation, though should work (albeit slowly) for a single CPU.

To run the first experiment, use the script run_expressivity_experiment.sh and 
pass in the following arguments:

        --times : number of instances per setup
        --min : minimum polynomial degree to test
        --max : maximum polynomial degree to test
        --bud : list of alpha values to test passes as a comma separated string 
        --reg_th: Accuracy threshold on SVM fit of initial classifier 
        --strat_thr: Accuracy thhreshold to determine k'

The multinomial_overfit class in expressivity_experiment.py also includes a 
method called "one_instance" which can be used to get the regular and strategic 
labels of a set of instances.

Because the second experiment requires fewer parameters, it can be directly run 
using the main function in accuracy_experiment.py. 
