Resorting to non-linear solutions
Linear models are approachable and interpretable, given the one-to-one relation between feature columns and regression coefficients. Sometimes, anyway, you may want to try non-linear solutions in order to check whether models that are more complex can model your data better and solve your prediction problem in a more expert manner. Support Vector Machines (SVMs) are an algorithm that rivaled neural networks for a long time and they are still a viable option thanks to recent developments in terms of random features for large-scale kernel machines (Rahimi, Ali; Recht, Benjamin. Random features for large-scale kernel machines. In: Advances in neural information processing systems. 2008. pp. 1177-1184). In this recipe, we demonstrate how to leverage Keras and obtain a non-linear solution to a classification problem.
Getting ready
We will still be using functions from the previous recipes, including define_feature_columns_layers and make_input_fn...
