Criterion

rusty_machine::learning::glm

Trait Criterion 

Source 
pub trait Criterion {
    type Link: LinkFunc;

    // Required method
    fn model_variance(&self, mu: f64) -> f64;

    // Provided methods
    fn initialize_mu(&self, y: &[f64]) -> Vec<f64> { ... }
    fn compute_working_weight(&self, mu: &[f64]) -> Vec<f64> { ... }
    fn compute_y_bar(&self, y: &[f64], mu: &[f64]) -> Vec<f64> { ... }
    fn apply_link_func(&self, vec: Vector<f64>) -> Vector<f64> { ... }
    fn apply_link_inv(&self, vec: Vector<f64>) -> Vector<f64> { ... }
    fn link_grad(&self, mu: f64) -> f64 { ... }
}
Expand description

The criterion for the Generalized Linear Model.

This trait specifies a Link function and requires a model variance to be specified. The model variance must be defined to specify the regression family. The other functions need not be specified but can be used to control optimization.

Required Associated Types§

The link function of the GLM Criterion.

Required Methods§

Source

fn model_variance(&self, mu: f64) -> f64

The variance of the regression family.

Provided Methods§

Source

fn initialize_mu(&self, y: &[f64]) -> Vec<f64>

Initializes the mean value.

By default the mean takes the training target values.

Source

fn compute_working_weight(&self, mu: &[f64]) -> Vec<f64>

Computes the working weights that make up the diagonal of the W matrix used in the iterative reweighted least squares algorithm.

This is equal to:

1 / (Var(u) * g’(u) * g’(u))

Source

fn compute_y_bar(&self, y: &[f64], mu: &[f64]) -> Vec<f64>

Computes the adjustment to the fitted values used during fitting.

This is equal to:

g`(u) * (y - u)

Applies the link function to a vector.

Applies the inverse of the link function to a vector.

Computes the gradient of the link function.

Implementors§