Optimizers as enums with internal state and methods

This chapter builds on the previous enum-based optimizer design. We now give each variant its own internal state and encapsulate behavior using methods. This pattern is useful when you want enum-based control flow with encapsulated logic.

Defining the optimizer enum

Each optimizer variant includes its own parameters and, when needed, its internal state.

#![allow(unused)]
fn main() {
#[derive(Debug, Clone)]
pub enum Optimizer {
    /// Gradient Descent optimizer with a fixed learning rate.
    GradientDescent { learning_rate: f64 },
    /// Momentum-based optimizer with velocity tracking.
    Momentum {
        learning_rate: f64,
        momentum: f64,
        velocity: Vec<f64>,
    },
}
}

Here, GradientDescent stores only the learning rate, while Momentum additionally stores its velocity vector.