Summary
In this chapter, we explored the different strategies for implementing SVC. We used linear SVC (which does not use kernels), which can perform very well when our classes are linearly separable. We then examined how to use the kernel trick to extend SVC to cases where the classes are not linearly separable. Finally, we used one-versus-one and one-versus-rest classification to handle targets with more than two values.
SVC is an exceptionally useful technique for binary and multiclass classification. It can handle both straightforward and complicated relationships between features and the target. There are few supervised learning problems for which SVMs should not at least be considered. However, it is not very efficient with very large datasets.
In the next chapter, we will explore another popular and flexible classification algorithm, Naive Bayes.
