As a part of the course work, we had to implement Pegasos and Adagrad SVM solver for Twitter data obtained.
This assumes a random starting point for the separating hyperplane and then adjusts the plane as mini-batches of data are sampled.
The update step is slightly different for the two algorithms and AdaGrad tends to converge faster because it steps much faster than Pegasos.
The two alogirthms were then tested on Testing Data and the error rates of the training and testing are reported.