Creating spline features
Linear models expect a linear relationship between the predictor variables and the target. However, we can use linear models to model non-linear effects if we first transform the features. In the Performing polynomial expansion recipe, we saw how we can unmask linear patterns by creating features with polynomial functions. In this recipe, we will discuss the use of splines.
Splines are used to mathematically reproduce flexible shapes. They consist of piecewise low-degree polynomial functions. To create splines, we must place knots at several values of x. These knots indicate where the pieces of the function join. Then, we fit low-degree polynomials to the data between two consecutive knots.
There are several types of splines, such as smoothing splines, regression splines, and B-splines. scikit-learn supports the use of B-splines to create features. The procedure to fit and, therefore, return the spline values for a certain variable, based on a polynomial...
