Class GeneralizedAdditiveModelOptions<T>
Configuration options for Generalized Additive Models (GAMs), which are flexible regression models that combine multiple simple functions to model complex relationships.
public class GeneralizedAdditiveModelOptions<T> : RegressionOptions<T>Type Parameters
TThe numeric type used for calculations (typically double or float).
- Inheritance
-
GeneralizedAdditiveModelOptions<T>
- Inherited Members
Remarks
Generalized Additive Models extend linear regression by allowing non-linear relationships between predictors and the target variable. They work by fitting a separate smooth function (typically a spline) for each predictor variable and then adding these functions together. This approach maintains much of the interpretability of linear models while allowing for more flexible relationships.
For Beginners: Think of a Generalized Additive Model as a more flexible version of linear regression. In linear regression, you assume each input feature has a straight-line relationship with your target variable (like y = mx + b from high school math). GAMs relax this assumption by allowing each feature to have its own curved relationship with the target.
For example, if you're predicting house prices, a linear model might assume that price increases by a fixed amount for each additional square foot. A GAM could capture that small houses might see a bigger price increase per square foot than very large houses, where additional space might add less value.
GAMs achieve this flexibility while still being relatively easy to interpret compared to "black box" models like neural networks, because you can visualize how each individual feature affects the prediction. They're a good middle ground between simple linear models and complex machine learning algorithms.
Properties
Degree
Gets or sets the degree of the polynomial splines used in the model.
public int Degree { get; set; }Property Value
- int
The polynomial degree, defaulting to 3 (cubic splines).
Remarks
This parameter determines the degree of the polynomial used for each spline segment. Higher degrees allow for smoother transitions between spline segments but can lead to more oscillations. The default value of 3 corresponds to cubic splines, which provide a good balance of smoothness and stability.
For Beginners: This setting controls how smooth the transitions are between the curve pieces in your model. With the default value of 3, the model uses cubic polynomials (like y = ax³ + bx² + cx + d) which create very smooth curves with continuous first and second derivatives (meaning the curve and its slope change gradually without sudden jumps).
Think of it like the difference between connecting dots with straight lines (degree 1), gentle curves (degree 2), or even smoother curves (degree 3). Higher degrees create smoother transitions but can sometimes lead to unexpected wiggles in regions with sparse data. Lower degrees create simpler curves but might have more abrupt changes in direction.
The most common choices are:
- 1: Linear splines (straight line segments)
- 2: Quadratic splines (parabolic segments)
- 3: Cubic splines (the default, very smooth)
For most applications, the default cubic splines (degree 3) work well, providing a good balance of smoothness and predictability.
NumSplines
Gets or sets the number of spline basis functions to use for each feature.
public int NumSplines { get; set; }Property Value
- int
The number of splines, defaulting to 10.
Remarks
This parameter controls how many spline basis functions are used to represent the smooth function for each predictor variable. More splines allow for more complex and wiggly functions, while fewer splines result in smoother functions with less flexibility.
For Beginners: This setting controls how flexible each curve in your model can be. With the default value of 10, the model uses 10 simple curve pieces (called "splines") that it combines to form a smooth curve for each feature. Think of it like drawing a curve using a limited number of control points - more points (higher NumSplines) let you create more complex curves with more twists and turns, while fewer points force the curve to be simpler and smoother.
If you set this too high (like 30+), your model might become too flexible and start fitting to noise in your data (overfitting). If you set it too low (like 3), your model might be too rigid to capture important patterns (underfitting). The default of 10 is a good starting point for many problems, but you might adjust it based on how complex you think the relationships in your data are and how much data you have (more data can support more splines).