Experiment hyperparameters that Lorraine tunes

[vxbrandon]

Do hyperparameters sweeps with hyperparameters that Lorraine tunes:

data_sizes = [100, 200, 1600]
val_props = [.0, .1, .25, .5, .75, .9]
num_neumann_terms in [1, 0, -1]
seed in [1, 2, 3]

Use the hyperparams_sweep.py file in the hyperparameters_sweep branch.

I looked at this and this. If anyone could find other hyperparameters that Lorraine tuned, please let me know!