All of the regressions that we have worked with up until now are parametric. They place restrictions on how the inputs can influence the response, for example forcing the relationship to act through a linear model. This is called parametric analysis because these models have parameters and you fit a model to the data by optimizing parameters.
Nonparametric regression algorithms make fewer assumptions about the relationship between x and y. In their purest form, these algorithms will learn the true relationship between x and y as you observe more data, regardless of what this truth looks like. As you accumulate data, your predictions ...