1 unstable release
0.1.0 | May 16, 2022 |
---|
#19 in #line-search
Used in 3 crates
(2 directly)
40KB
586 lines
Line search, also called one-dimensional search, refers to an optimization procedure for univariable functions.
Available algorithms
- MoreThuente
- BackTracking
- BackTrackingArmijo
- BackTrackingWolfe
- BackTrackingStrongWolfe
References
- Sun, W.; Yuan, Y. Optimization Theory and Methods: Nonlinear Programming, 1st ed.; Springer, 2006.
- Nocedal, J.; Wright, S. Numerical Optimization; Springer Science & Business Media, 2006.
Examples
use line::linesearch;
let mut step = 1.0;
let count = linesearch()
.with_initial_step(1.5) // the default is 1.0
.with_algorithm("BackTracking") // the default is MoreThuente
.find(5, |a: f64, out: &mut Output| {
// restore position
x.veccpy(&x_k);
// update position with step along d
x.vecadd(&d_k, a);
// update value and gradient
out.fx = f(x, &mut gx)?;
// update line search gradient
out.gx = gx.vecdot(d);
// update optimal step size
step = a;
// return any user defined data
Ok(())
})?;
let ls = linesearch()
.with_max_iterations(5) // the default is 10
.with_initial_step(1.5) // the default is 1.0
.with_algorithm("BackTracking") // the default is MoreThuente
.find_iter(|a: f64, out: &mut Output| {
// restore position
x.veccpy(&x_k);
// update position with step along d
x.vecadd(&d_k, a);
// update value and gradient
out.fx = f(x, &mut gx)?;
// update line search gradient
out.gx = gx.vecdot(d);
// update optimal step size
step = a;
// return any user defined data
Ok(())
})?;
for success in ls {
if success {
//
} else {
//
}
}
Dependencies
~9–20MB
~281K SLoC