Function linfa::composing::platt_scaling::platt_newton_method [−][src]
pub fn platt_newton_method<'a, F: Float, O>(
reg_values: ArrayView1<'a, F>,
labels: ArrayView1<'a, bool>,
params: &PlattParams<F, O>
) -> Result<(F, F), PlattNewtonResult>
Run Newton’s method to find optimal A
and B
values
The optimization process happens in two steps, first the closed-form Hessian matrix and gradient vectors are calculated. Then a line-search tries to find the optimal learning rate for each step.