Machine Learning with Swift

Model training, evaluation, and selection

Usually, it is better to start from simple and classical models because sometimes the simplest model performs the best. But this is just a rule of thumb, not a law of nature.

Every machine learning algorithm embodies some assumptions or prior knowledge about the data: KNN assumes that similar examples are of the same class, linear regression assumes linear dependencies and normally distributed errors, many models assume independence or limited dependencies between features or samples, and so on. This helps them to generalize behind the training data successfully. All these assumptions work only because the samples are not distributed uniformly across the space of all possible inputs, and there is ...