CallCVX.varyparam: Simple R interface to CVX to solve sequence of problems.

In jacobbien/CVXfromR: Matlab and CVX from R

Description

This function takes a string containing what you'd normally put between cvx_begin and cvx_end. Returns CVX output. Allows a sequence of problems to be performed that are identical except that a single scalar parameter is varied.

Usage

CallCVX.varyparam(cvx.code, const.vars, tuning.param, opt.var.names, 
setup.dir = NULL, norun=FALSE, matlab.call = "matlab", cvx.modifiers = NULL)

Arguments

cvx.code	String containing call to CVX, i.e. what's inside cvx_begin and cvx_end
const.vars	List of non-optimization variables used in CVX expression. Labels of list elements should match what the corresponding variable name in cvx.code. E.g. for the Lasso, this would be list(y=y, x=x, lam=lam, p=p)
tuning.param	A list with a single vector (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. Each element of this vector is a level of the parameter. The problem will be solved at each such level.
opt.var.names	Array of names of optimization variables. E.g. for the Lasso, this would be "beta"
setup.dir	Directory containing the file cvx_setup. If not needed, leave NULL.
norun	Default FALSE. Mostly for debugging purposes. Doesn't call Matlab. Returns the command that would be run in Matlab.
matlab.call	How Matlab can be invoked through the "system" command. Default: "matlab" but even if this is the alias in your default shell, "system" might use a different shell in which "matlab" is not recognized.
cvx.modifiers	Optional string of modifiers passed to CVX on same line as cvx_begin. E.g. "quiet" or "sdp".

Details

This function is very similar to CallCVX, but allows one to solve a sequence of problems in which only one (scalar) non-optimization variable is varied. The intended use is for when a problem has a tuning parameter, and you want to solve the problem at various values of this parameter. Using this function is much more efficient than wrapping CallCVX in a loop.

Value

A list of the optimization variables specified in opt.var.names. Each variable is a list of length equal to the number of problems solved. Also,

cvx_optval	The optimal values as returned by CVX in each of the problems.
time	Elapsed total time specifically for all the CVX calls (i.e. excludes opening Matlab, loading data, etc.)

Author(s)

Jacob Bien

References

M. Grant and S. Boyd. CVX: Matlab software for disciplined convex programming, version 1.21. http://cvxr.com/cvx, April 2011.

M. Grant and S. Boyd. Graph implementations for nonsmooth convex programs, Recent Advances in Learning and Control (a tribute to M. Vidyasagar), V. Blondel, S. Boyd, and H. Kimura, editors, pages 95-110, Lecture Notes in Control and Information Sciences, Springer, 2008. http://stanford.edu/~boyd/graph_dcp.html.

See Also

CallCVX

Examples

# Solve the lasso at multiple lambda values using CVX
n <- 50
p <- 10
x <- matrix(rnorm(n * p), n, p)
beta <- rnorm(p)
y <- x %*% beta + 0.1 * rnorm(n)
lam <- seq(2, 0.1, length=10)
# to call CVX, set setup.dir to be the directory containing "cvx_setup.m"
## Not run: 
setup.dir <- "change/this/to/your/cvx/directory"
lasso <- CallCVX.varyparam("variables beta(p);minimize(square_pos(norm(y-x*beta,2))/2+lam*norm(beta,1))",
                 const.vars=list(p=p, y=y, x=x),
		 tuning.param=list(lam=lam),
                 opt.var.names="beta", setup.dir=setup.dir)

## End(Not run)

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