evclust (version 2.0.3)

harris: Harris gradient-based optimization algorithm

Description

The optimization algorithm implemented in harris is described on Silva & Almeida (1990) and summarized in Denoeux & Masson (2004). The four parameters are:

options[1]

Display parameter : 1 (default) displays some results.

options[2]

Maximum number of iterations (default: 100).

options[3]

Relative error for stopping criterion (default: 1e-4).

options[4]

Number of iterations between two displays.

Usage

harris(fun, x, options = c(1, 100, 1e-04, 10), tr = FALSE, ...)

Value

A list with three attributes:

par

The minimizer of fun found.

value

The value of fun at par.

trace

The trace, a list with two attributes: 'time' and 'fct' (if tr==TRUE).

Arguments

fun

Function to be optimized. The function 'fun' should return a scalar function value 'fun' and a vector 'grad' containing the partial derivatives of fun at x.

x

Initial value (a vector).

options

Vector of parameters (see details).

tr

If TRUE, returns a trace of objective function vs CPU time

...

Additional parameters passed to fun

Author

Thierry Denoeux.

References

F. M. Silva and L. B. Almeida. Speeding up backpropagation. In Advanced Neural Computers, R. Eckmiller, ed., Elsevier-North-Holland, New-York, 151-158, 1990.

T. Denoeux and M.-H. Masson. EVCLUS: Evidential Clustering of Proximity Data. IEEE Transactions on Systems, Man and Cybernetics B, Vol. 34, Issue 1, 95--109, 2004.

See Also

pcca

Examples

opt<-harris(function(x) return(list(fun=sum(x^2),grad=2*x)),rnorm(2),tr=TRUE)
print(c(opt$par,opt$value))
plot(opt$trace$fct,type="l")