Add SVM loss

[jpsember]

http://cs231n.github.io/linear-classify/#svm

[jpsember]

This is what our function should look like.

We should have a custom SVM loss node that is able to dynamically determine loss (and gradient) based upon the training sample's label, yi.

[jpsember]

Let's attempt this first with the 'toy example':

http://cs231n.github.io/neural-networks-case-study/

[jpsember]

• [x] Allow iterating over training data, accumulating gradient into sum

[jpsember]

Results agree with toy example from lecture notes:

[jpsember]

This is the function, including the SVM cost and regularization loss: