To visualize the results, a comparison of two two-dimensional cases for the algorithms trained on the first 300 of the training examples are displayed below, along with their training errors and in addition for logistic regression the number of iterations it took Newton’s Method to converge. The plots on the left are perimeter versus texture and the plots on the right are texture versus radius. Note the decision boundaries and for both versions of GDA and the contours of each multivariate gaussian modeling the positive and negative classes.