Minimise f over a convex set by gradient steps followed by projection.
morie_pgdsdg is the module alias of projected_gradient_descent.
Usage
morie_pgdsdg(f, grad, x0, project, ...)
Arguments
- f
Objective function.
- grad
Its gradient.
- x0
Starting point.
- project
Projection onto the feasible set.
- ...
Further arguments: step size, rule, iteration and tolerance controls; see the source.
Value
A list with the iterate, objective value and convergence trace.
Examples
f <- function(v) sum((v - c(2, -1))^2)
g <- function(v) 2 * (v - c(2, -1))
res <- morie_pgdsdg(f, g, x0 = c(0, 0), project = project_nonneg)
res$x
#> [1] 2 0
