sgd-vs-gd In these experiments we trying to check if it better to train SGD from one starting point or GD from different?