Function opt

src/opt.rs:238–264

Minimizes an objective using gradient descent. The objective `f` is minimized using a standard gradient descent algorithm. The argument `fd` must return the values of the derivatives for each parameter and is executed in each iteration for the current parameters. The argument `init` contains the initial parameters and `opts` contains the options for the gradient descent algorithm. <script type="

(f: &O, fd: &D, init: &[f64], opts: OptParams<f64>)

/// # }
/// ```
pub fn opt<O, D>(f: &O, fd: &D, init: &[f64], opts: OptParams<f64>) -> OptResult<f64>
    where O: Fn(&[f64]) -> f64, D: Fn(&[f64]) -> Vec<f64> {

    let alpha = opts.alpha.unwrap_or(0.1);
    let iter = opts.iter.unwrap_or(1000);
    let eps = opts.eps;

    let mut r = vec![];
    let mut p = init.to_vec();
    let mut stopped = false;

    for _ in (0..iter) {
        let i = p.sub(&fd(&p).mul_scalar(alpha));
        r.push((i.clone(), f(&i)));
        stopped = eps.is_some() && i.iter().zip(p.iter()).all(|(&x, &y)| num::abs(x - y) <= eps.unwrap());
        p = i;
        if stopped {
            break;
        }
    }

    OptResult {
        params: p.to_vec(),
        fvals: r,
        stopped: stopped
    }
}

// TODO duplicated code
pub fn opt_hypothesis(h: &Hypothesis, x: &Matrix<f64>, y: &[f64], opts: OptParams<f64>) -> OptResult<f64> {