callconvex.varyparams: Simple R interface to Convex.jl to solve sequence of problems

In jacobbien/convexjulia: Call Convex.jl from R

Description

Like callconvex but allows a sequence of problems to be performed that are identical except that a single parameter is varied. The parameter varied can be scalar or vector valued.

Usage

callconvex.varyparams(opt.vars, pr.def, const.vars, vary.param, opt.var.names,
  pr.solve = "solve!(pr)", code.before = "", code.after = "",
  norun = FALSE, julia.call = "julia", delete.temp = TRUE)

Arguments

opt.vars	string of code to declare optimization variables
pr.def	string of code defining a problem called pr
const.vars	list of non-optimization variables used in expression. Labels of list elements should be the names of the variables.
vary.param	list with a single vector or matrix (with a name). E.g. list(lam=c(0.1, 1, 2)). This is a non-optimization variable used in CVX expression that you want varied. If a vector, then each element is a parameter; if a matrix, then each column is a vector-parameter. The problem will be solved at each such level.
opt.var.names	array of names of optimization variables to return.
pr.solve	string of code for solving problem pr. This can be to specify the solver used, for example.
code.before	optional string of julia code to run before solving problems
code.after	optional string of julia code to run after solving problems
norun	Default FALSE. Mostly for debugging purposes. Returns the command that would be run in julia without it actually opening julia.
julia.call	How julia can be invoked through the system command. Default: "julia". However, even if this is the alias in your default shell, system might use a different shell in which "julia" is not recognized. For example, in OS X this might be something like "/Applications/Julia-0.3.3.app/Contents/Resources/julia/bin/julia"
delete.temp	Default TRUE. Indicates whether the temporary files that are created should be deleted.

Details

This function uses calljulia.

Value

Returns optval and status from Convex.jl in addition to optimal values of variables named in opt.var.names. If optimization variable is a vector, then returns a matrix in which each column corresponds to a solution; if optimization variable is a matrix, then each returns a 3-dimensional array whose third mode corresponds to different solutions.

References

Udell, Madeleine, et al. "Convex optimization in Julia." Proceedings of the 1st First Workshop for High Performance Technical Computing in Dynamic Languages. IEEE Press, 2014.

Convex.jl julia package. https://github.com/cvxgrp/Convex.jl

Grant, Michael, Stephen Boyd, and Yinyu Ye. Disciplined convex programming. Springer US, 2006.

See Also

callconvex

Examples

## Not run: 
set.seed(123)
n <- 200; p <- 50
x  <- matrix(rnorm(n * p), n, p)
beta <- rnorm(p)
y <- x %*% beta + 0.1 * rnorm(n)
julia <- "/Applications/Julia-0.3.3.app/Contents/Resources/julia/bin/julia"
# vary lambda for the lasso (with an unpenalized intercept)
pr.def <- "pr = minimize(sumsquares(y - b0 - x * b) + lam * norm(b, 1))"
opt.vars <- "b = Variable(p); b0 = Variable(1)"
lasso <- callconvex.varyparams(opt.vars = opt.vars, pr.def = pr.def,
                              const.vars = list(x = x, y = y, p = p),
                              vary.param = list(lam = seq(9, 2, length=10)),
                              opt.var.names = c("b", "b0"),
                              julia.call = julia)

## End(Not run)

jacobbien/convexjulia documentation built on May 18, 2019, 8:01 a.m.