Train a linear SVM with sequential SGD updates on hinge loss, L2 weight regularization, and signed class predictions.